Category: news

Z-Image Prompt Assist
Z-Image Prompt Assist

Benchmark comparison chart of all aesthetics offered via “PASS” and how Z-IMAGE failed most of them.

comfyui

generative ai

Comparison

This is a comparison benchmark running Z-Image Turbo model testing the various Genre options found in my latest toll Prompt Assistant v1.1, affectionately nicknamed PASS.
For the sake of testing purposes, I am using exactly the same prompt, same seed, the same sampler, same scheduler and same resolution of 824 x 1328 ( 1536 x 2048 would have been ideal ). The only change will be on the initial aesthetic value that will be note on each image caption.
I hope this comparision can be somewhat useful to understand what genre means in generative art. You are invited to play around with the PASS tool and use to guide your future generations with JSon and Long string text outputs ready to be copied and pasted nto your workflow.

Access PASS (free use) – Prompt Assistant v1.1

Z-Image using DMPP_SDE + BETA

Z-Image using EULER_ANCESTRAL + DDIM UNIFORM
Prompt to be used

Photorealistic, Bright / Airy aesthetic. Subject: A 40yr female Mixed Latino-European minimalist (wearing a matte grey suit without shirt underneath, long trousers and high-heels), pose: leaning on her car, intense gaze expression. Featuring: a Ferrari (yellow with pristine body conditions). Scene set in a skyscrapper rooftop in Dubai with soft diffused light. Technical: high angle, shot on Hasselblad medium format, 50mm lens.
WHAT I LEARNED FROM THIS BENCHMARK

Z-Image Turbo excels in beautiful photorealistic images of daily subjects as long it stays within a cinematic / editorial portraiture realm.
When experimenting with more artistic and edgy prompts pushing the aesthetics far away from the common reality, Z-Image Turbo starts to fail and ignores these prompt instructions, as if it is addicted or trained only to a certain aesthetic only.
Z-Image can not follow specific Aesthetic/Genre prompts.

Z-IMAGE BEST SAMPLER + SCHEDULER RESULTS

DDIM + DDIM_UNIFORM

DMPP_SDE + BETA

DMPP_SDE + KARRAS

Sampler + scheduler comparison – all native combinations

DDIM + SIMPLE

DDIM + SGM_UNIFORM

DDIM + KARRAS

DDIM + EXPONENTIAL

DDIM + BETA

DDIM + NORMAL

DDIM + DDIM_UNIFORM

DDIM + LINEAR_QUADRATIC

DDIM + KL_OPTIMAL

EULER + SIMPLE

EULER + SGM_UNIFORM

EULER + KARRAS

EULER + EXPONENTIAL

EULER + BETA

EULER + NORMAL

EULER + DDIM_UNIFORM

EULER + LINEAR_QUADRATIC

EULER + KL_OPTIMAL

EULER ANCESTRAL + SIMPLE

EULER ANCESTRAL + SGM_UNIFORM

EULER ANCESTRAL + KARRAS

EULER ANCESTRAL + EXPONENTIAL

EULER ANCESTRAL + BETA

EULER ANCESTRAL + NORMAL

EULER ANCESTRAL + DDIM_UNIFORM

EULER ANCESTRAL + LINEAR_QUADRATIC

EULER ANCESTRAL + KL_OPTIMAL

DMPP_SDE + SIMPLE

DMPP_SDE + SGM_UNIFORM

DMPP_SDE + KARRAS

DMPP_SDE + EXPONENTIAL

DMPP_SDE + BETA

DMPP_SDE + NORMAL

DMPP_SDE + DDIM_UNIFORM

DMPP_SDE + LINEAR_QUADRATIC

DMPP_SDE + KL_OPTIMAL

Bright / Airy

Moody

Dark

Soft / Dreamy

Ethereal

Analog / Film Grain

Vintage

Retro

Nostalgic

Vaporwave

Synthwave

Cyberpunk

Futuristic

Dystopian

Utopian

Minimalist

Maximalist

Clean

Grunge

Gothic

Luxury

Cozy

Street Style

Whimsical

Surreal

Psychedelic

Glitch

Y2K

Dark Academia

Light Academia

Neon

Pastel

Monochrome

Just aesthetics is not enough

The benchmark proved that by changing just the aesthetics value will cause no identifiable changes to the output result. Z-Image can not understand the aesthetic value on it’s own, rather you have to change completely one portion of the prompt to ensure the aesthetic is described in more words.
Take a closer look on the next examples with comparison on aesthetics word value and whole portion prompt rewriting.
Important note: Styles like surreal, glitch, dystopian and others similar that pushes a very specific appeal to the end result will absolutely not be considered by Z-Image rendering this model useless if your intent is anything other than Editorial / Street / Cinematic or Photorealistic.

Neon

Neon + Extra prompting

Final veridict

By no means I am assuming that Z-Image Turbo has severe flaws, rather pin-pointing that this model seems to be built to attend a specific use: realism for editorial, cinematic and street photography featuring humans.
Rendering objects can be complex for Z-Image and logos and specific parts details on objects can not be handled properly. For example: the vehicle is identifiable as a Ferrari but the logo, emblem and details on the wheels, handles and front lights are completely mishandled.
Some of the details flaws can be rebuilt with upscaling techniques or even inpainting, thus, requiring an extra process to fix it.
02/05/2026
Short Guide on Generating Perfect Product Imagery with AI for Small Business

Short Guide on Generating Perfect Product Imagery with AI for Small Business

Using AI to generate the perfect shot or product placement of your product

comfyui

generative ai

Product visualization

The goal for any local or small-scale retailer is to produce content that performs two distinct roles: it must stop the scroll on social media and answer technical questions on the product page.
For many small business owners, the primary barrier to digital growth isn’t the quality of the product—it’s the quality of the content. Capturing the rich texture of a fine leather purse or the subtle sheen of hardware requires a level of photographic skill, studio lighting, and post-production that is often financially out of reach for a lean operation. Without these professional visuals, even the most premium items can appear “homemade” on a storefront, which can inadvertently signal a lack of credibility and lead to lower conversion rates.

Artificial Intelligence has fundamentally changed this dynamic, democratizing high-end production for local retailers. Instead of requiring a physical studio and a professional crew, business owners can now use AI as a digital art director and editor. By leveraging generative models, you can take a basic, well-lit photo from your smartphone and transform it into a sophisticated asset library. AI solves the perennial challenges of staging, lighting, and perspective, allowing small brands to present a visual identity that rivals global luxury houses at a fraction of the cost.

From Smartphone to Storefront, build Brand Narrative and Lifestyle Context one Step at a time

iPhone shot of the product taken by a person with no photography expertise

1. Starting point

For platforms like Instagram and TikTok, the goal is “aspirational storytelling.” The product needs to look like it belongs in the hands of your ideal customer.
Using various AI models and workflows, we can take that basic iPhone photo and transplant the purse into a high-end, realistic setting without losing the integrity of the original leather texture, features, dimensions, details and other characteristics.
◄ Take the example here on the left: the purse was shot on iPhone using basic camera settings and natural lighting. The picture itself is not bad, would suffice for Marketplace or other online selling markets, but certainly is not good enough for a social media ad or e-commerce presentation.

2. The German, the Chinese and the American Models

The original image has to have some correction done. By adding some dynamic characteristics to the main element we can better grab the attention of the audience. But how/which AI Model to use?
In order to accomplish that, we have to remove the background, change the camera lens length, maybe even change the angle a little bit. Adding more context to the image means choosing a scenario and a story to tell.
As of January 2026 there are three main AI Edit Models competitors on the market: Flux Kontext is the German AI model, Qwen Edit is the Chinese AI model and Gemini Nano Banana is the popular American AI model. All of them have pros and cons which let us adjust through a prompt all the changes we need. Take a look at the comparison pictures below using the same edit requests.
Click images to see larger version.

Edit generated with Flux Kontext – German model

Edit generated using Qwen – Chinese model

Edit generated using Gemini – American model

Although all three results looks good, they are far beyond something professional. The dimension of the purse is affected and even the background looks distorted or way too much artificial. The overall mood, light technique and camera angle seems good, but placing the product requires more steps. For that, we are going to add extremely targeted details in the prompting.

All of them need obvious adjusts, including the purse dimensions that are way too off.

Edited with Flux – German model

Purse is smaller – Qwen Edit – Chinese version

Futuristic background is removed and changed to a night scene in Tokyo – Gemini – American version

3. Reshooting the product for professional usage

Original iPhone shot

High-quality, professional images are rewarded by search engine algorithms. By using AI to generate clean, high-resolution catalog shots, you improve your chances of appearing in Google Image Search. When you name these files descriptively, you are actively building an SEO foundation that brings local customers directly to your door.

Edit using DEIS Sampler

Edit using EULER Sampler

Edit using DMPP_2M Sampler

The “levitating” approach for this product goes hand in hand with its premise. This is why we are going to make further edits and use this idea as our main central point to promote this product.

Original product shot on iPhone

+

Background image on sale on iStockPhoto

All of them are loosing quality as new generations are made. Once the right posing is found, we have to take it to an upscale model.

The product was placed on the chosen background and all the merge details were taken care of by the AI model

Added a pair of plants to contrast and counterbalance the composition, so it does not look flat and even.

Changed product position and added text element

Changed logo texture

Upscaled final image ready to be used for a social media ad

16/01/2026
Z-Image Comparison Test
Z-Image Comparison test

benchmark on model shift / steps / scheduler

Just a few days after Flux 2 hiting the AI scene on end November 2025, the chinese came along and shove both feet at the door with this Qwen based fast model called Z-Image that excels in realistic portraits, celebrities and landmarks.

This my personal comparison chart to benchmark its features and this is my attempt to understand what each node does and what each settings affects the generated result. Follow along to check it for yourself too.

Seed = fixed
Steps = see comparison below
CFG = 1 (will adhere strictly to the prompt, without deviations)
Sampler = res_multistep (suggested by the model’s author)
Scheduler = simple (suggested by the model’s author)
Denoise = 1
No LORAS, no extra nodes
Timestamps are based on a GeForce 3080TI 12 GB RAM
Positive Prompt below / Negative Prompt = none
steps

Steps means how many times the latent space will be calculated before delivering the final output. Z-Image is a fast model and was developed to use less steps. Here, I found the best results are between 8 and 12. Less steps generates too few details, more steps generates extra noise and some inconsistencies, mainly on the hair.

Latina female with thick wavy hair, blinking to the camera, there are some out of focus tree leaves in front of her. Breezy seaside light, warm tones, cinematic close-up.
Steps 4
Time 7″

Steps 8
Time 12″

Steps 12
Time 18″

Steps 24
Time 26″

Steps 32
Time 44″

Steps 50
Time 68″
Model shift

Model Shift node means how noise or how polished the output will be. The lower the value, more artifacts will be present, the greater the value, more waxy the result. The model’s author recommends using Model Shift value = 3. All generations below were using 9 steps.

Blonde female with thick wavy hair, is sending a cute kiss to the camera, there are some out of focus tree leaves in front of her. Breezy seaside light, warm tones, cinematic close-up.
Model Shift 0.2
Time 14″

Model Shift 0.5
Time 14″

Model Shift 1
Time 14″

Model Shift 2
Time 18″

Model Shift 3
Time 14″

Model Shift 2
Time 18″

Model Shift 30
Time 14″

Model Shift 100
Time 18″
Denoise

Noise means how soon must the model’s latent space calculations stop before processing the final output. The model’s author recommends using Noise value = 1. All generations below were using 9 steps and shift 3. Very interesting to see how noise greatly affects the output.

Black female with thick wavy hair, is spontaneously smiling to the camera, there are some out of focus tree leaves in front of her. Breezy seaside light, warm tones, cinematic close-up.
Denoise 1
Time 14″

Denoise 0.9
Time 14″

Denoise 0.8
Time 14″

Denoise 0.7
Time 13″

Denoise 0.6
Time 13″

Denoise 0.5
Time 13″
Sampler

Sampler means the agent technology that will interpret the latent space according to the scheduler model and convert to a visual output. Each agent has it’s own algorithm and will influence mainly the time it needs to generate the final output, with some minor subtle aspect changes. The model’s author recommends using sampler = res_multistep. More noticeable are the light streaks on her arm and the out of focus leaves. All generations below were using 9 steps and shift 3.

Brazilian female with thick wavy hair, is holding her surf board, there are some out of focus tree leaves in front of her. Breezy seaside light, warm tones, cinematic close-up.
Sampler res_multistep
Time 19″

Sampler euler
Time 13″

Sampler heun
Time 24″

Sampler dmpp_2m
Time 14″

Sampler ipndm
Time 13″

Sampler deis
Time 13″

Sampler ddim
Time 13″

Sampler res_5s
Time 64″

Sampler res_6s_ode
Time 76″
03/12/2025
RES4LYF Sampler Comparison Chart
RES4LYF Sampler Comparison Chart

All the ClownsharKSampler compared side by side

comfyui

generative ai

Comparison

Since RES4LYF was released and gained popularity for having almost a hundred samplers using ClownSharkSampler, I decided to test each one of them to the same prompt and resolution in an attempt to find the most sharp and realistic generative render.
Positive promtp used:

In a futuristic world, there is this beautiful woman wearing heavy make-up in geometric shapes. She is standing in front of a neutral vibrant yellow background, while the dynamic city lights casts high contrasting light on her face and mechanical parts. This is a portrait shot on Leica M6 using 85 mm lens. The skin texture is very sharp to the pore-level and has some imperfections. Film grain. fluxlisimo_cinematic

Official workflow JSON: download from GitHub

Model: flux1-dev-fp8.safetensors
Clip 1: clip_l.safetensors
Clip 2: t5xxl_fp16.safetensors
VAE: ae.safetensors
LORA: 3d_illustration + fluxlisimo_cinematic + grain_scape
Dimension: 640 x 640
Flux Conditioning: 1.5
Fixed Seed: 347746757495977
Steps: 16
eta: 0.8
Scheduler: normal
Rendered using GTX 3080
Veredict

Fastest
Exponential DDIM and Linear Euler in 24,5 seconds

Veredict

Best quality and prompt adeherence
Linear rk6_7s in 167 seconds

Veredict

Very good result I
Fully implicit radau_iia_7s in 227,5 seconds

Veredict

Very good result II
Exponential Res_4_Minchev in 95 seconds

All samplers compared side to side

Below you will find each Sampler Algorithm Method followed by the sampler’s name and the time (in seconds) it took to generate/render a 640 x 640 px image on a GTX 3080 10GB.

Overall selected samplers for good resulting

At the end of this Article you will find some generations using the Samplers I am electing as the best out-coming results. These generations were rendered at 24 steps and at 512 x 1024 px on a GTX 3080 10GB. Jump directly to this section.

Multistep

Res_2m, Res_3m, Dpmpp_2m, Dpmpp_3m, Abnorsett_2m, Abnorsett_3m, Abnorsett_4m, Deis_2, Deis_3m, DEis_4m

multistep
res_2m
29”

multistep
res_3m
36”

multistep
dpmpp_2m
29”

multistep
dpmpp_3m
36”

multistep
abnorsett_2m
29”

multistep
abnorsett_3m
32”

multistep
abnorsett_4m
42”

multistep
deis_2m
28”

multistep
deis_3m
36”

multistep
deis_4m
47”

Exponential

Res2, Res3, Res4, Res5, res6, Res8, Res10, Res 15, Res16, etdrk2, etdrk3, dpmpp, lawson, ddim

exponential
res_2s
47”

exponential
res_2s_stable
47,5”

exponential
res_2s_rkmk2e
47”

exponential
res_3s
70”

exponential
res_3s_non-monotonic
69,5”

exponential
res_3s_alt
70,5”

exponential
res_3s_cox_matthews
70”

exponential
res_3s_lie
71”

exponential
res_3s_sunstar
71,5”

exponential
res_3s_strehmel_weinner
92,5”

exponential
res_3s_krogstadt
92,5”

exponential
res_4s_cox_matthews
93,5”

exponential
res_4s_cfree4
93”

exponential
res_4s_friedli
106”

exponential
res_4s_minchev
95”

exponential
res_4s_munthe-kaas
91”

exponential
res_5s
116”

exponential
res_5s_hochbruck-ostermann
115”

exponential
res_6s
140”

exponential
res_8s
187”

exponential
res_10s
239”

exponential
res_15s
366”

exponential
res_16s
384”

exponential
etdrk2_2s
46”

exponential
etdrk3_a_3s
69”

exponential
etdrk3_b_3s
69”

exponential
etdrk4_4s
93,5”

exponential
etdrk3_4s_alt
92”

exponential
dpmpp_2s
47,5”

exponential
dpmpp_sde_2s
47”

exponential
dpmpp_3s
70”

exponential
lawson_2a_2s
48”

exponential
lawson_2b_2s
48”

exponential
lawson4_4s
93”

exponential
lawson41-gen_4s
93”

exponential
lawson41-gen-mod_4s
92”

exponential
ddim
24,5”

Hybrid

Pec423, pec 433, abnorsett, lawson

hybrid
pec423_2h3s
47”

hybrid
pec433_2h3s
70”

hybrid
abnorsett2_1h2s
48”

hybrid
abnorsett3_2h2s
47,5”

hybrid
lawson42-gen-mod_1h4s
93”

hybrid
lawson42-gen-mod_2h4s
94”

hybrid
lawson42-gen-mode_3h4s
94”

Linear

ralston, midpoint, heun, houwen-wray, kutta, ssprk3, RK, bogacki_shampine, dormand-prince, tsi, euler

linear
ralston_2s
47,5”

linear
ralston_3s
71”

linear
ralston_4s
93”

linear
midpoint_2s
47”

linear
heun_2s
48,5”

linear
heun_3s
71”

linear
houwen-wray_3s
71,5”

linear
kutta_3s
71”

linear
ssprk3_3s
71”

linear
ssprk4_4s
110”

linear
rk38_4s
95”

linear
rk4_4s
93”

linear
rk6_7s
167”

linear
bogacki_shampine_4s
95”

linear
bogacki_shampine_7s
165”

linear
dormand-prince_6s
141”

linear
dormand-prince_13s
318,5”

linear
tsi_7s
167”

linear
euler
24,5”

Diagonal implicit

irk_exp_diag, kraaijevanger_spijkjer, qin_zhang, pareschi_russo, crouzeix

diagonal implicit
irk_exp_diag_2s
74”

diagonal implicit
kraaijevanger_spijkjer_2s
74”

diagonal implicit
qin_zhang_2s
74”

diagonal implicit
pareschi_russo_2s
73”

diagonal implicit
pareschi_russo_alt_2s
76”

diagonal implicit
crouzeix_2s
74”

diagonal implicit
crouzeix_3s
100”

diagonal implicit
crouzeix_3s_alt
102”

Fully implicit

gauss_legendre, RADAU, lobatto

fully implicit
gauss_legendre_2s
74”

fully implicit
gauss_legendre_3s
101”

fully implicit
gauss_legendre_4s
130”

fully implicit
gauss_legendre_5s
160”

fully implicit
gauss_legendre_5s_ascending
159”

fully implicit
radau_ia_2s
73”

fully implicit
radau_ia_3s
76”

fully implicit
radau_iia_2s
74”

fully implicit
radau_iia_3s
99,5”

fully implicit
radau_iia_3s_alt
100”

fully implicit
radau_iia_5s
164”

fully implicit
radau_iia_7s
227,5”

fully implicit
radau_iia_9s
303”

fully implicit
radau_iia_11s
375”

fully implicit
lobatto_iiia_2s
54”

fully implicit
lobatto_iiia_3s
80”

fully implicit
lobatto_iiia_4s
109,5”

fully implicit
lobatto_iib_2s
54”

fully implicit
lobatto_iib_3s
77,5”

fully implicit
lobatto_iib_4s
103,5”

fully implicit
lobatto_iic_2s
50,5”

fully implicit
lobatto_iic_3s
78,5”

fully implicit
lobatto_iic_4s
106,5”

fully implicit
lobatto_iic_star_2s
51”

fully implicit
lobatto_iic_star_3s
78”

fully implicit
lobatto_iid_2s
51”

fully implicit
lobatto_iid_3s
77,5”

… end of the sampler’s list …

Overall good results

RES_4S_MINCHEV, DEIS_3M, RES_2M, RK6_7S, lobatto_iiib_4s, lobatto_iiic_4s

exponential
res_4s_minchev
24 steps
150”

exponential
res_4s_minchev
24 steps
151.5”

exponential
res_4s_minchev
24 steps
151,5”

exponential
res_4s_minchev
24 steps
157”

multistep
deis_3m
24 steps
60”

multistep
deis_3m
24 steps
60”

multistep
deis_3m
24 steps
59”

multistep
deis_3m
24 steps
59”

multistep
res_2m
24 steps
51”

multistep
res_2m
24 steps
51”

multistep
res_2m
24 steps
50”

multistep
res_2m
24 steps
51”

linear
rk6_7s
24 steps
275”

linear
rk6_7s
24 steps
169”

linear
rk6_7s
24 steps
169”

linear
rk6_7s
24 steps
296”

fully implicit
lobatto_iiib_4s
24 steps
169”

fully implicit
lobatto_iiib_4s
24 steps
169”

fully implicit
lobatto_iiib_4s
24 steps
175”

fully implicit
lobatto_iiib_4s
24 steps
173,5”

fully implicit
lobatto_iiic_4s
24 steps
170”

fully implicit
lobatto_iiic_4s
24 steps
167”

fully implicit
lobatto_iiic_4s
24 steps
174”

fully implicit
lobatto_iiic_4s
24 steps
166,5”

End of the first study and batch comparison test.

Workflow: {“id”: “72f6123a-4c3f-4675-9e45-86722ffa8aeb”, “revision”: 0, “last_node_id”: 22, “last_link_id”: 33, “nodes”: [{“id”: 13, “type”: “VAEDecode”, “pos”: [859.5225830078125, 217.15052795410156], “size”: [140, 46], “flags”: {}, “order”: 8, “mode”: 0, “inputs”: [{“name”: “samples”, “type”: “LATENT”, “link”: 12}, {“name”: “vae”, “type”: “VAE”, “link”: 13}], “outputs”: [{“name”: “IMAGE”, “type”: “IMAGE”, “links”: [14, 16]}], “properties”: {“cnr_id”: “comfy-core”, “ver”: “0.3.44”, “Node name for S&R”: “VAEDecode”}, “widgets_values”: [], “color”: “#322”, “bgcolor”: “#533”, “shape”: 1}, {“id”: 10, “type”: “CLIPTextEncode”, “pos”: [115.9236831665039, 642.4409790039062], “size”: [400, 200], “flags”: {“collapsed”: true}, “order”: 3, “mode”: 0, “inputs”: [{“name”: “clip”, “type”: “CLIP”, “link”: 7}], “outputs”: [{“name”: “CONDITIONING”, “type”: “CONDITIONING”, “links”: [9]}], “title”: “Negative Prompt”, “properties”: {“cnr_id”: “comfy-core”, “ver”: “0.3.44”, “Node name for S&R”: “CLIPTextEncode”}, “widgets_values”: [“”], “color”: “#322”, “bgcolor”: “#533”, “shape”: 1}, {“id”: 14, “type”: “SaveImageExtended”, “pos”: [1060, 210], “size”: [231.4296875, 718], “flags”: {}, “order”: 9, “mode”: 0, “inputs”: [{“name”: “images”, “type”: “IMAGE”, “link”: 14}, {“name”: “positive_text_opt”, “shape”: 7, “type”: “STRING”, “link”: null}, {“name”: “negative_text_opt”, “shape”: 7, “type”: “STRING”, “link”: null}], “outputs”: [], “properties”: {“cnr_id”: “save-image-extended-comfyui”, “ver”: “2.64.0”, “Node name for S&R”: “SaveImageExtended”}, “widgets_values”: [“RES4LYF_”, “sampler_name, cfg, steps, %F %H-%M-%S”, “comparisionbatch”, “”, “-“, “disabled”, false, “”, true, 4, “last”, true, true, “.jpg”, 90], “shape”: 1}, {“id”: 1, “type”: “ClownModelLoader”, “pos”: [-203.28634643554688, 222.3040313720703], “size”: [270, 266], “flags”: {}, “order”: 0, “mode”: 0, “inputs”: [], “outputs”: [{“name”: “model”, “type”: “MODEL”, “links”: [33]}, {“name”: “clip”, “type”: “CLIP”, “links”: [7, 28]}, {“name”: “vae”, “type”: “VAE”, “links”: [10, 13]}], “properties”: {“cnr_id”: “RES4LYF”, “ver”: “81fe5f4d02c1a86d4cbb1ef92c652055f574810a”, “Node name for S&R”: “ClownModelLoader”}, “widgets_values”: [“flux1-dev-fp8.safetensors”, “default”, “clip_l.safetensors”, “t5xxl_fp16.safetensors”, “.none”, “.none”, “flux”, “ae.safetensors”], “shape”: 1}, {“id”: 11, “type”: “FluxGuidance”, “pos”: [-211.47117614746094, 563.1766357421875], “size”: [270, 58], “flags”: {}, “order”: 1, “mode”: 0, “inputs”: [{“name”: “conditioning”, “type”: “CONDITIONING”, “link”: null}], “outputs”: [{“name”: “CONDITIONING”, “type”: “CONDITIONING”, “links”: null}], “properties”: {“cnr_id”: “comfy-core”, “ver”: “0.3.44”, “Node name for S&R”: “FluxGuidance”}, “widgets_values”: [1.5]}, {“id”: 22, “type”: “Lora Loader Stack (rgthree)”, “pos”: [120, 90], “size”: [271.8785095214844, 246], “flags”: {}, “order”: 4, “mode”: 0, “inputs”: [{“name”: “model”, “type”: “MODEL”, “link”: 33}, {“name”: “clip”, “type”: “CLIP”, “link”: 28}], “outputs”: [{“name”: “MODEL”, “type”: “MODEL”, “links”: [32]}, {“name”: “CLIP”, “type”: “CLIP”, “links”: [29]}], “properties”: {“cnr_id”: “rgthree-comfy”, “ver”: “944d5353a1b0a668f40844018c3dc956b95a67d7”, “Node name for S&R”: “Lora Loader Stack (rgthree)”}, “widgets_values”: [“3d-illustration.safetensors”, 0.5000000000000001, “fluxlisimo_cinematic_v1_FLUX.safetensors”, 0.8500000000000002, “GrainScape.safetensors”, 0.5000000000000001, “None”, 0], “color”: “#233”, “bgcolor”: “#355”, “shape”: 1}, {“id”: 9, “type”: “CLIPTextEncode”, “pos”: [111.22306060791016, 391.9552917480469], “size”: [418.7828674316406, 200], “flags”: {}, “order”: 6, “mode”: 0, “inputs”: [{“name”: “clip”, “type”: “CLIP”, “link”: 29}], “outputs”: [{“name”: “CONDITIONING”, “type”: “CONDITIONING”, “links”: [8]}], “title”: “Positive Prompt”, “properties”: {“cnr_id”: “comfy-core”, “ver”: “0.3.44”, “Node name for S&R”: “CLIPTextEncode”}, “widgets_values”: [“In a futuristic world, there is this beautiful woman wearing heavy make-up in geometric shapes. She is standing in front of a neutral vibrant yellow background, while the dynamic city lights casts high contrasting light on her face and mechanical parts. This is a portrait shot on Leica M6 using 85 mm lens. The skin texture is very sharp to the pore-level and has some imperfections. Film grain. fluxlisimo_cinematic”], “color”: “#232”, “bgcolor”: “#353”, “shape”: 1}, {“id”: 15, “type”: “PreviewImage”, “pos”: [1320, 200], “size”: [508.1442565917969, 708.9977416992188], “flags”: {}, “order”: 10, “mode”: 0, “inputs”: [{“name”: “images”, “type”: “IMAGE”, “link”: 16}], “outputs”: [], “properties”: {“cnr_id”: “comfy-core”, “ver”: “0.3.44”, “Node name for S&R”: “PreviewImage”}, “widgets_values”: [], “shape”: 1}, {“id”: 19, “type”: “Note”, “pos”: [840, 530], “size”: [210, 88], “flags”: {}, “order”: 2, “mode”: 0, “inputs”: [], “outputs”: [], “properties”: {}, “widgets_values”: [“347746757495977n347746757495977”], “color”: “#432”, “bgcolor”: “#653”, “shape”: 1}, {“id”: 12, “type”: “VAEEncodeAdvanced”, “pos”: [116.57431030273438, 690.6495361328125], “size”: [270, 278], “flags”: {}, “order”: 5, “mode”: 0, “inputs”: [{“name”: “image_1”, “shape”: 7, “type”: “IMAGE”, “link”: null}, {“name”: “image_2”, “shape”: 7, “type”: “IMAGE”, “link”: null}, {“name”: “mask”, “shape”: 7, “type”: “IMAGE”, “link”: null}, {“name”: “latent”, “shape”: 7, “type”: “LATENT”, “link”: null}, {“name”: “vae”, “shape”: 7, “type”: “VAE”, “link”: 10}], “outputs”: [{“name”: “latent_1”, “type”: “LATENT”, “links”: null}, {“name”: “latent_2”, “type”: “LATENT”, “links”: null}, {“name”: “mask”, “type”: “MASK”, “links”: null}, {“name”: “empty_latent”, “type”: “LATENT”, “links”: [11]}, {“name”: “width”, “type”: “INT”, “links”: null}, {“name”: “height”, “type”: “INT”, “links”: null}], “properties”: {“cnr_id”: “RES4LYF”, “ver”: “81fe5f4d02c1a86d4cbb1ef92c652055f574810a”, “Node name for S&R”: “VAEEncodeAdvanced”}, “widgets_values”: [“false”, 512, 1024, “red”, false, “4_channels”], “color”: “#322”, “bgcolor”: “#533”, “shape”: 1}, {“id”: 4, “type”: “ClownsharKSampler_Beta”, “pos”: [553.3905639648438, 220.2578887939453], “size”: [270, 662], “flags”: {}, “order”: 7, “mode”: 0, “inputs”: [{“name”: “model”, “shape”: 7, “type”: “MODEL”, “link”: 32}, {“name”: “positive”, “shape”: 7, “type”: “CONDITIONING”, “link”: 8}, {“name”: “negative”, “shape”: 7, “type”: “CONDITIONING”, “link”: 9}, {“name”: “latent_image”, “shape”: 7, “type”: “LATENT”, “link”: 11}, {“name”: “sigmas”, “shape”: 7, “type”: “SIGMAS”, “link”: null}, {“name”: “guides”, “shape”: 7, “type”: “GUIDES”, “link”: null}, {“name”: “options”, “shape”: 7, “type”: “OPTIONS”, “link”: null}], “outputs”: [{“name”: “output”, “type”: “LATENT”, “links”: [12]}, {“name”: “denoised”, “type”: “LATENT”, “links”: null}, {“name”: “options”, “type”: “OPTIONS”, “links”: null}], “properties”: {“cnr_id”: “RES4LYF”, “ver”: “81fe5f4d02c1a86d4cbb1ef92c652055f574810a”, “Node name for S&R”: “ClownsharKSampler_Beta”}, “widgets_values”: [0.7000000000000002, “multistep/res_2m”, “normal”, 28, -1, 1, 1.0000000000000002, 586933988910742, “randomize”, “standard”, true], “color”: “#223”, “bgcolor”: “#335”, “shape”: 1}], “links”: [[7, 1, 1, 10, 0, “CLIP”], [8, 9, 0, 4, 1, “CONDITIONING”], [9, 10, 0, 4, 2, “CONDITIONING”], [10, 1, 2, 12, 4, “VAE”], [11, 12, 3, 4, 3, “LATENT”], [12, 4, 0, 13, 0, “LATENT”], [13, 1, 2, 13, 1, “VAE”], [14, 13, 0, 14, 0, “IMAGE”], [16, 13, 0, 15, 0, “IMAGE”], [28, 1, 1, 22, 1, “CLIP”], [29, 22, 1, 9, 0, “CLIP”], [32, 22, 0, 4, 0, “MODEL”], [33, 1, 0, 22, 0, “MODEL”]], “groups”: [], “config”: {}, “extra”: {“ds”: {“scale”: 1.2884080000000004, “offset”: [-62.504483003015366, -106.68595565476912]}, “frontendVersion”: “1.23.4”, “VHS_latentpreview”: false, “VHS_latentpreviewrate”: 0, “VHS_MetadataImage”: true, “VHS_KeepIntermediate”: true}, “version”: 0.4, “seed_widgets”: {“4”: 7}}

RES4LYF ClownsharKSampler
Comparison chart generated on 19/20 July 2025
18/07/2025
New Flux 1 Kontext Dev free to use workflow and examples

New Flux 1 Kontext Dev free to use workflow and examples

Flux Kontext was only available as a paid service through several online APIs but on June 2025 the free version was released under the name of FLUX 1 Kontext Dev and can be run locally, aka, using Comfy UI or Diffusers.
What Kontext does is what most of us from the AI community were waiting for over the last years: generate changes to images while maintaining consistency. And that’s a game-changer in the history of generative AI.
See below some practical examples where I noted the Sampler technology used as well.
For a complete installation guide and basic workflows, refer to the official page here: https://docs.comfy.org/tutorials/flux/flux-1-kontext-dev

Original image generated
with Fluxmania

Kontext “opened hazel eyes” generated
with Deis sampler

Kontext “opened hazel eyes” generated
with DMPP 2M sampler

Original image generated
with Fluxmania and GrainScape LORA

Kontext “tall grass field in Scottland” generated
with IPNDM sampler

Kontext “golden wheat crops” generated
with IPNDM sampler

Kontext “paradisiac beach in Philipines” generated
with IPNDM sampler

Original image generated
with Fluxmania and GrainScape LORA

Kontext “social suit, white shirt and long pants” generated
with IPNDM sampler

Kontext “brazilian soccer uniform” generated
with IPNDM sampler

It is completely stunning how Kontext can handle the contextualization of its InPaint method and reading the whole image to a degree that can understand depth of field, noise, camera angle, character unique marks, posture, color grading and more.
Important to note that the engine Flux Kontext Dev provided by Black Forest Labs, the original author, does not allow NSFW, nudity, erotic or pornographic generations.
The downside is that Kontext not yet can easily change facial expressions. When trying expressions such as angry, surprised or sensual, it does not comply. Changing the Sampler might fix that. See example below.

Original image generated
with Flux 1 Dev

Kontext “surprised” or “angry” generated
with any Sampler returns always with error

Kontext “smile” generated
with IPNDM sampler

Kontext “looking in awe with mouth open” generated
with Euler sampler

Commercial application for real world use

While playing around and messing with prompting is fun and leads to visually stunning results, there must have a more serious and commercial application point of view, otherwise, the technology would render useless for businesses.
This is where the workflow of Kontext Dev lets us combine two images together and generates a single output with perfect matching colors and composition.
See the examples below.

Real photo of the product taken from Fashion Nova

Fluxmania Checkpoint generated
female feet with Feet Model LORA

Kontext combined generated
with Euler sampler

Real photo of the product taken from Fashion Nova

Fluxmania Checkpoint generated
female feet with Feet Model LORA

Kontext combined generated
with DPMPP 2M sampler

Real photo of the product taken from Fashion Nova

Fluxmania Checkpoint generated
female feet with Feet Model LORA

Kontext combined generated
with DPMPP 2M sampler

The results are not yet optimal because the sandal is rendered twice with the same strap position, on the left and on the right feet. Plus, it still somehow has something way too artificial to be used commercially. Maybe with fine prompting or stacking some LORAs to the workflow could generate better results.
As take more time to study I will publish my findings on this website and LinkedIn.

02/07/2025
Optimizing LinkedIn profile for growth and performance
Optimizing LinkedIn profile for growth and performance
📌 Step 1: Post Consistently (Content Strategy)
🎯 Post 3-5x per week (mix of formats):
✔ Text Posts (Thought leadership, questions, opinions)
✔ Images/Carousels (Tips, stats, infographics)
✔ Videos (Short-form, behind-the-scenes, value-driven)
✔ Articles/News (Comment on industry trends)
🔹 Best Times to Post:
Weekdays (Tue-Thu), 8-10 AM or 6-8 PM (local time).
Use LinkedIn Scheduler (native tool) or Buffer/Hootsuite.

Tool: Crystal
Analyzes LinkedIn connections to suggest communication styles.
Helps tailor your profile & outreach messages.

Tool: Peekalink
Generates AI-powered LinkedIn headlines and summaries.

Tool: Canva (AI features)
Auto-generates professional banners and profile pics.

Tool: Jasper (formerly Jarvis)
Writes LinkedIn posts, captions, and articles in seconds.

Tool: OpusPro
Creates short videos from long content (great for LinkedIn).

Tool: Hypefury
Optimizes post timing & captions based on engagement data.

Tool: Loom
Records quick video updates to humanize your profile.

📌 Step 2: Engage Authentically (Build Relationships)
💬 Daily Engagement Routine:
1️⃣ Like & Comment on top posts in your niche (add value, not just “Great post!”).
2️⃣ Join & Engage in LinkedIn Groups (share insights, answer questions).
3️⃣ Reply to Comments on your own posts (boosts engagement).
4️⃣ Tag relevant people (when appropriate, but don’t overdo it).

Tool: Expandi
Sets up automated LinkedIn drip campaigns (e.g., follow-up messages after connecting).

Tool: Dux-Soup
Automates profile visits & personalized connection requests.

Tool: PhantomBuster
Scrapes leads & automates likes/comments (use ethically).
📌 Step 3: Grow Your Network Strategically
🔍 Targeted Connection Strategy:
✅ Connect with:
Industry peers (alumni, conference attendees).
Potential clients/leads (follow & engage before requesting).
Influencers & thought leaders (comment on their posts first).
🚫 Avoid:
Random bulk requests (LinkedIn may flag you).
Spammy messages (“Can I promote my product?”).
🔹 Personalize Connection Requests:
“Hi [Name], I noticed we both work in [industry] and loved your post on [topic]. Would love to connect!”

Tool: MeetAlfred
Automates connection requests & follow-ups with personalization.

Tool: Prospeo
Finds targeted leads based on job titles, companies, etc.

Tool: Crystal
Suggests personalized outreach messages for each connection.

📌 Step 4: Leverage LinkedIn Features
✨ Use These Tools to Stand Out:
📌 Newsletters – Repurpose long-form posts into a weekly newsletter.
📌 LinkedIn Live – Host quick Q&As or industry updates.
📌 Polls & Stories – Engage your audience interactively.
📌 SEO Optimization – Use keywords in your profile & posts (helps with discovery).

Tool: Repurpose.io
Turns YouTube videos into LinkedIn carousels/posts.

Tool: OpusClip
Auto-cuts long videos into viral clips for LinkedIn.

Tool: Jasper
Repurposes blog posts into LinkedIn threads.
📌 Step 5: Track & Improve (Analytics)
📊 Check Weekly Metrics:
✔ Post Engagement (Likes, Comments, Shares).
✔ Profile Views (Are people visiting your page?).
✔ Follower Growth (Are you gaining quality connections?).
🔹 Double Down on What Works:
If carousels perform well, create more.
If video gets more comments, post more videos.

Tool: Vaetas
Tracks LinkedIn post performance and suggests improvements.

Tool: Vaetech Suite
Automates analytics reporting for LinkedIn growth.

Tool: SocialBu
Aggregates engagement metrics across platforms (LinkedIn included).

📌 Step 7: Monetize & Offer Value (Long-Term Growth)
💰 Turn LinkedIn Presence into Opportunities:
📌 Lead Magnet – Offer a free guide/checklist in your bio (use Linktree).
📌 Consulting/Services – Showcase case studies & testimonials.
📌 Affiliate/Products – Promote relevant offers subtly.
🎯 Your 1-Week Action Plan
Day 1-2: Optimize profile (photo, headline, about section).
Day 3-4: Post 3x this week (text + image + video).
Day 5-7: Engage daily (like/comment on 10+ posts, send 5 connection requests).

By following these steps consistently, you’ll see steady growth in visibility, engagement, and opportunities.

🎯 Quick-Start AI Workflow
Optimize Profile → Crystal + Canva
Post 3x/Week → Jasper + Hypefury
Engage Daily → Dux-Soup (limited use)
Grow Network → MeetAlfred + Prospeo
Track Success → Vaetas + SocialBu
18/06/2025

Flux Dev vs Fluxmania

Flux 1 (dev) vs Flux GGUF vs Fluxmania

I recently came across to people telling wonders about Fluxmania, a fine-tuned legacy checkpoint from Flux and available at Civiat.ai. The prunned model has 22,5GB but quantized versions are also available through GGUF formats.

This is my personal testings with direct comparision on Flux 1 Dev against its own GGUF version against the Fluxmania Checkpoint.

I am using Sampler Euler + Scheduler SGM_Unigform on 24 steps and CFG at 1.0 and same seed for every generation.

				
					Close-up portrait of a woman with long black hair partially obscured by a glowing, motion-blurred sword held vertically in front of her face, dramatic light streaks crossing composition, low contrast, cinematic high dynamic range, highly detailed, intricate chainmail headpiece and septum ring, modern warrior aesthetic blending medieval armor with cyberpunk fashion, neon glow effects, futuristic armor textures, ambient lighting, dark background, film grain, sharp focus, contrasting cool and warm tones, moody atmosphere, detailed facial features, soft rim lighting, subtle lens flares, high-resolution realistic rendering, cinematic composition, depth of field, subtle reflections and metallic textures, dynamic motion blur, editorial style, hyper-realistic,

Flux 1 Dev
774726078651743

				
					Close-up portrait of a woman with long black hair partially obscured by a glowing, motion-blurred sword held vertically in front of her face, dramatic light streaks crossing composition, low contrast, cinematic high dynamic range, highly detailed, intricate chainmail headpiece and septum ring, modern warrior aesthetic blending medieval armor with cyberpunk fashion, neon glow effects, futuristic armor textures, ambient lighting, dark background, film grain, sharp focus, contrasting cool and warm tones, moody atmosphere, detailed facial features, soft rim lighting, subtle lens flares, high-resolution realistic rendering, cinematic composition, depth of field, subtle reflections and metallic textures, dynamic motion blur, editorial style, hyper-realistic,

Flux 1 Dev Q8 GGUF
774726078651743

				
					Close-up portrait of a woman with long black hair partially obscured by a glowing, motion-blurred sword held vertically in front of her face, dramatic light streaks crossing composition, low contrast, cinematic high dynamic range, highly detailed, intricate chainmail headpiece and septum ring, modern warrior aesthetic blending medieval armor with cyberpunk fashion, neon glow effects, futuristic armor textures, ambient lighting, dark background, film grain, sharp focus, contrasting cool and warm tones, moody atmosphere, detailed facial features, soft rim lighting, subtle lens flares, high-resolution realistic rendering, cinematic composition, depth of field, subtle reflections and metallic textures, dynamic motion blur, editorial style, hyper-realistic,

Fluxmania
774726078651743

				
					Low angle shot of a high-speeding Lancia Delta on a street crowded with street market under cloudy skies, diffused lighting, muted colors, film grain, sharp focus, contrasting cool and warm tones, moody atmosphere, soft lighting and shadows, subtle lens flares, high-resolution realistic rendering, dynamic composition, subtle reflections and metallic textures, dynamic motion blur, editorial style

Flux 1 Dev
911677283233535

				
					Low angle shot of a high-speeding Lancia Delta on a street crowded with street market under cloudy skies, diffused lighting, muted colors, film grain, sharp focus, contrasting cool and warm tones, moody atmosphere, soft lighting and shadows, subtle lens flares, high-resolution realistic rendering, dynamic composition, subtle reflections and metallic textures, dynamic motion blur, editorial style

Flux 1 Dev Q8 GGUF
911677283233535

				
					Low angle shot of a high-speeding Lancia Delta on a street crowded with street market under cloudy skies, diffused lighting, muted colors, film grain, sharp focus, contrasting cool and warm tones, moody atmosphere, soft lighting and shadows, subtle lens flares, high-resolution realistic rendering, dynamic composition, subtle reflections and metallic textures, dynamic motion blur, editorial style

Fluxmania
911677283233535

				
					Eye level angle shot of Lancia Delta 4S rally posing parked in the middle of a crowded street under direct sunlight daylight with long casting shadows, vivid colors, film grain, sharp focus, contrasting cool and warm tones, 35mm Kodak Vision3 250D, vivid colors, cinematic dynamic range, moody atmosphere, subtle lens flares, high-resolution realistic rendering, dynamic composition, subtle reflections and metallic textures, dynamic motion blur, editorial style

Flux 1 Dev
27081982270819

				
					Eye level angle shot of Lancia Delta 4S rally posing parked in the middle of a crowded street under direct sunlight daylight with long casting shadows, vivid colors, film grain, sharp focus, contrasting cool and warm tones, 35mm Kodak Vision3 250D, vivid colors, cinematic dynamic range, moody atmosphere, subtle lens flares, high-resolution realistic rendering, dynamic composition, subtle reflections and metallic textures, dynamic motion blur, editorial style

Flux 1 Dev Q8 GGUF
27081982270819

				
					Eye level angle shot of Lancia Delta 4S rally posing parked in the middle of a crowded street under direct sunlight daylight with long casting shadows, vivid colors, film grain, sharp focus, contrasting cool and warm tones, 35mm Kodak Vision3 250D, vivid colors, cinematic dynamic range, moody atmosphere, subtle lens flares, high-resolution realistic rendering, dynamic composition, subtle reflections and metallic textures, dynamic motion blur, editorial style

Fluxmania
27081982270819

				
					Editorial photography full body of a woman wearing an all-white minimalist outfit while posing sitting on a decorative moon in a studio backdrop simulating outter space. Studio light softbox lighting, professional product photography. natural skin texture. Kodak Portra 400, soft pastel tones, natural skin tones.

Flux 1 Dev
764863394705616

				
					Editorial photography full body of a woman wearing an all-white minimalist outfit while posing sitting on a decorative moon in a studio backdrop simulating outter space. Studio light softbox lighting, professional product photography. natural skin texture. Kodak Portra 400, soft pastel tones, natural skin tones.

Flux 1 Dev Q8 GGUF
764863394705616

				
					Editorial photography full body of a woman wearing an all-white minimalist outfit while posing sitting on a decorative moon in a studio backdrop simulating outter space. Studio light softbox lighting, professional product photography. natural skin texture. Kodak Portra 400, soft pastel tones, natural skin tones.

Fluxmania
764863394705616

15/06/2025

Flux Schnell Q2 vs Q6 GGUF

Flux Schnell Q2 vs Q6 GGUF

Q2

Q6

deis – normal

deis – normal

deis – beta

deis – beta

deis – sgm uniform

deis – sgm uniform

dpmm2 – beta

dpmm2 – beta

dpmm2 – exponential

dpmm2 – exponential

dpmm2 – karras

dpmm2 – karras

dpmm2 – normal

dpmm2 – normal

25/04/2025
How Brands Are Targeting Gen-Z with Reshaped Marketing Strategies

How Brands Are Targeting Gen-Z with Reshaped Marketing Strategies

Brands have to evolve and be flexible in order to stay relevant. This means trying out something never done before.
The most controversial and difficult generation to retain attention and convince has to be the Gen-Z for it is still being studied and for they have access to all information all the time, they are a not dumb generation as many think, but rather informed, maybe not by the official news media, but certainly informed.
When dealing with Gen-Z the most important point is that brands and companies are not dealing with the consumer alone, but with a culture.
Sony once famous for saying “We don’t follow the market, we create it.” in the 80s and 90s, sure needs to reevaluate that today.

broadway lowers its nose

Broadway is undergoing a brand transformation. Long seen as a destination for upscale, older audiences, New York City’s theaters are now finding fresh appeal with Gen Z through influencer marketing and more accessible pricing strategies.
This means their are investing in viral Tik Tok strategies to reach their target audience and had to repurpose the pricing lowering it to a more affordable and acceptable rate by this audience. You can buy tickets online around US$ 50 on their official website.

LOWE’S and ea sports, inside video game advertising strategy

American home improvement retailer LOWE’S is betting their marketing strategy in partnership with giant EA Sport Entertainment.
They will have an inside video game advertising trying to reach target audience while playing NFL Madden ’25.
The concept art above shows how the brand will be displayed inside the game as LOWE’S is already an NFL sponsor. Lowe’s is positioning the EA deal as an extension of its broader sports marketing strategy, but with a particular focus on digitally savvy young consumers. The company has started to reverse slumping sales but the home improvement category faces challenges amid a mounting trade war.
Typical to video games partnerships, the tie-up is rooted in a need to reach Gen Z and young millennial shoppers who have made the hobby into a powerhouse of mainstream entertainment. Those cohorts are perceived as particularly fickle when it comes to brand loyalty, and Lowe’s is putting a spotlight on its MyLowe’s Rewards platform through the EA Sports franchises.
You can read the original article at Marketing Dive.

14/04/2025
Live Portrait AI

Live Portrait AI

This is LivePortrait AI and it lets you track a recorded face video or even real-time recording face, to animate a photo.

I was playing around with this version and it’s still glitchy and funny, but in no time, it will be reliable enough for commercial and professional use.
Official website: liveportrait.org/

18/02/2025