Diffus
Get started
models/PixelWave - FLUX.1-dev 03

PixelWave - FLUX.1-dev 03

humblemikey
|
7/15/2025
|
4:40:46 AM
| Discussion|
0
A woman in a black dress reading a book on a dark red velvet sofa in a cozy, dimly lit library room with a lit fireplace, floral painting, and tall bookshelves.
Daytime photo of a Los Angeles suburban neighborhood showing a white SUV parked on the street in front of houses and palm trees.
Young woman in short jeans and white tanktop leaning against a 1950s red car on a colorful Cuban street in Havana with pastel tenement buildings.
A scenic aerial view of a Roman city set in a hilly Italian landscape with dry dirt paths in the foreground, mountains in the background, and the ocean to the left under a clear blue sky.

Recommended Parameters

samplers

Euler Normal, dpmpp 2m sgm uniform, dpmpp_2s_ancestral, euler, DPM++ SDE, 2M SDE Karras, DPM++ SDE Karras, DPM++ 2M

steps

4 - 40

cfg

2.5 - 7

resolution

1024x1024, 1536x1536

vae

ae.safetensors - unknown

Recommended Hires (High-Resolution) Parameters

upscale

1.5 - null

Tips

You can use more steps to improve finer details, but the output doesn't change much after 8 steps.

If you want a neater/cleaner output, try increasing the guidance scale (CFG).

Mentioning a style in the prompt can help the model generate better results.

Try adding the upscale latent by node, and scale the latent by 1.5 to generate higher resolution images.

Avoid using 'quality' terms like 4K, 8K, masterpiece, high def, high quality unless necessary; can cause images to look overcooked.

For photographic styles, avoid terms like 'vibrant, intense, bright, high contrast, neon, dramatic' if you want a natural look.

Training was done with kohya_ss/sd-scripts using pagedlion8bit optimizer.

Freezing 'time_in', 'vector_in' and modulation parameters stops 'de-distillation'.

Avoid training single blocks over 15; set training blocks in FLUX section.

Learning rate 5e-6 trains fast but stop after a few thousand steps to avoid block corruption.

Version Highlights

Fine tuned for 5 weeks on my 4090.

Creator Sponsors

Model also available at: RunDiffusion and Runware.ai

Huge thank you to RunDiffusion for sponsoring the compute that made training this model possible!

For API access, partner with Runware.ai

PixelWave FLUX.1-schnell 04 - Apache 2.0!

Safetensor Files: 💾BF16 💾FP8 💾bnb FP4

GGUF Files: 💾Q8_0 🤗Q6_K 💾Q4_K_M

Links to 🤗VAE 🤗T5xxl 🤗CLIP L

Model also available at: RunDiffusion and Runware.ai

PixelWave FLUX.1 schnell version 04 is an aesthetic fine tune of FLUX.1-schnell. The training images were hand picked to ensure the model has a bias to eye catching images, with beautiful colors, textures and lighting.

  • Trained on the original schnell model, so Apache 2.0 license!

  • No special requirements to run. Supports FLUX LoRAs

  • Euler Normal, 8 steps.

You can use more steps to improve finer details, but the output doesn't change much after 8 steps.

Shout out to RunDiffusion

Huge thank you to RunDiffusion (co-creators of Juggernaut) for sponsoring the compute that made training this model possible! Figuring out how to train schnell without de-distilling the model required a lot of experimenting, and being able to utilize RunDiffusion's cloud compute made it a lot easier.

For those needing API access for this model, we're partnering with Runware.ai

I have made the FLUX.1-dev 04 version exclusive to RunDiffusion and Runware for the time being. When I release version 05 in future, I plan to release the dev 04 open weights.

Grateful for their support in getting this model out there, please check them out!

Training

Training was done with kohya_ss/sd-scripts. You can find my fork of Kohya here , which also contains changes to the sd-scripts submodule, make sure you clone both.

Use the fine tuning tab. I found the best results with the pagedlion8bit optimizer which also could run on my 4090 GPU 24GB. I found other optimizers struggle to learn anything.

I have frozen the time_in, vector_in and mod/modulation parameters. This stops the 'de-distillation'.

I avoid training single blocks over 15. You can set which blocks to train in the FLUX section.

LR 5e-6 trains fast, but you have to stop after a few thousand steps as it starts to corrupt blocks and slow down learning.

You can then block merge with an earlier checkpoint, replacing the corrupt blocks, and then continue training further.

Signs of corrupt blocks: paper texture over most images, loss of background details.

Contact

For business or commercial inquiries please reach out to us at pixelwave@rundiffusion.com. Licensing flux fine tunes. Customer training projects. Commercial AI development. The team can do it all!

PixelWave Flux.1-dev 03 fine tuned!

Safetensor Files: 💾BF16 💾FP8 💾NF4

GGUF Files: 💾Q8_0 🤗Q6_K 💾Q4_K_M

Links to 🤗VAE 🤗T5xxl 🤗CLIP L

The 'diffusers' files are actually the Q8_0 and Q4_K_M GGUF versions. GGUF files also available on huggingface.

I fine tuned version 03 from base FLUX.1-dev for over 5 weeks on my 4090. It is able to do different art styles, photography, and anime. Trick I discovered to help with LoRAs.

I used dpmpp 2m sgm uniform 30 steps for the showcase images. If you want a neater/cleaner output, try increasing the guidance. Also mentioning a style can help, so the model doesn't have to guess.

I also recommend try adding the upscale latent by node, and scale the latent by 1.5, e.g. generating an image that is 1536x1536 instead of 1024x1024.

PixelWave Flux.1-schnell 03

Safetensor Files: 💾FP8 💾NF4

GGUF Files: go to huggingface

I used dpmpp 2m sgm uniform 8 steps for the showcase images.

You can start with 4 steps, but there are less errors with anatomy if you run with more steps.

PixelWave Flux.1-dev 02

Safetensor Files: 💾BF16 💾FP8

GGUF Files: 💾Q8_0 🤗Q6_K 💾Q4_K_M

Version 02 has greatly improved black and dark images, and more reliable outputs with fewer issues with hands.

I recommend using dpmpp_2s_ancestral, beta, 14 steps. Or euler, simple, 20 steps.

Comfyui-GGUF Nodes

PixelWave 11 SDXL. A general purpose fine tuned model. Great for art and photo styles.

I use 20 steps, DPM++ SDE, CFG 4 to 6 or 40 steps, 2M SDE Karras

Accelerated Version - 5+ Steps, DPM++ SDE Karras, 2.5 CFG

PAG Recommended⚡Recommend 1.5 Scale, with CFG 3. Link to workflow

🔗Link to Expanded Gallery 🖼️

⭐Link to prompting guide.⭐ You don't need to use 'quality' terms such as 4K, 8K, masterpiece, high def, high quality, etc. Unless you want it, I recommend not using words such as 'vibrant, intense, bright, high contrast, neon, dramatic' for photographic styles if you a wanting a more natural look. This can cause images to look 'overcooked', but it's just the CLIP following your prompt. 🙂 If you do want vibrant, neon photos PixelWave will provide!

The focus for version 10 was to train the CLIP models, which improves the reliability, ensures you can produce a wide variety of styles, and better at following prompts.

Thanks to my friends who helped test: masslevel, blink, socalguitarist, klinter, wizard whitebeard.

Guide: Upscaling Prompts with LM Studio and Mikey Nodes

Guide: Add more details to your image using the skip step method

No need for the refiner model.

This model is not a mix of other models.

I also created Mikey Nodes which contains a lot of useful nodes. You can install it through comfy manager.

Contributor

humblemikey avatar

humblemikey

Previous
Artify´s Fantastic Flux Landscape Lora - artilands_v1.0
Next
Project0* - v4.0 Art_Realism_FP8
Use this model

Model Details

Model type

Checkpoint

Base model

Flux.1 D

Model version

FLUX.1-dev 03

Model hash

49b429b788

Creator

humblemikey's avatar
humblemikey

Discussion

Please log in to leave a comment.

Images by PixelWave - FLUX.1-dev 03

A woman in a black dress reading a book on a dark red velvet sofa in a cozy, dimly lit library room with a lit fireplace, floral painting, and tall bookshelves.
Daytime photo of a Los Angeles suburban neighborhood showing a white SUV parked on the street in front of houses and palm trees.
Young woman in short jeans and white tanktop leaning against a 1950s red car on a colorful Cuban street in Havana with pastel tenement buildings.
A scenic aerial view of a Roman city set in a hilly Italian landscape with dry dirt paths in the foreground, mountains in the background, and the ocean to the left under a clear blue sky.

anime Images

Anime-style Egyptian female kneeling in desert, with tan skin, white hair, red eyes, wearing a white robe and Egyptian collar, sunbeam in background.
Anime girl with short hair standing in a blue flower field at night, with a glowing blue light and galaxy in the sky, city lights in the distance.
A stylized anime girl with long black hair and striking green eyes, wearing a black outfit and thigh-high stockings adorned with red flowers tattoo, holding a katana against a red background.
A dramatic steampunk battle scene with airships looming over a city skyline, robots and steam-powered exosuits fighting on the street while an explosion lights up the center under stormy skies.
A smiling fox girl with blonde hair, yellow eyes, fox ears, and a fox tail, wearing a colorful kimono with fur trim and traditional hair ornaments, standing in front of a torii gate.
Anime girl with purple-blue hair in a pink china dress sits on a teal couch holding trays of drinks in a cafe setting.
Anime-style blonde girl wearing a white sundress and large sunhat with a sunflower decoration, standing happily in a sunflower field at sunset with her arms raised.
A confident woman with short white hair and piercing blue eyes wearing a sleeveless black turtleneck sweater, black pants, and fingerless gloves, standing in a dimly lit urban alleyway under a starry night sky.
Blonde woman wearing black and gold helldiver armor with cleavage cutout, thighboots, and holding a helmet, with blue eyes and a light smile
Anime girl with red hair in a braid, wearing a navy school uniform with gold trim, sitting with legs spread, looking at viewer with red eyes, and wearing red ribbons and earrings.

base model Images

Photorealistic scene of undead characters including zombies and skeletons walking through a spooky cemetery illuminated by glowing jack-o'-lanterns under a dark, ominous sky.

digital art Images

A translucent holographic spaceship model displayed inside a transparent cube in the middle of a bustling futuristic city square with tall buildings and digital billboards.
Photorealistic digital illustration of a futuristic bedroom featuring a holographic projection of a blue-toned space-themed character above the ornate bed, with a woman in a sleek futuristic outfit sitting on the bed in a luxurious room with blue walls and cityscape view.
A sci-fi female character with a long black ponytail and bangs, wearing a skin-tight white and emerald bodysuit, shown from behind with emphasis on her figure in comic style.
A glowing Starbucks coffee mug emitting violet and cyan sparks and steam in a dark setting.
Digital painting of a large dark bird with jagged feathers and outstretched wings, set against a textured intense red background with a bright glowing light on the right.
Digital portrait of a red-haired girl with green eyes and freckles on her face and shoulders, wearing a dark tank top against a dark background.
Anthro black cat wearing a stylish plaid suit sitting at a wooden desk, holding a glass of brown gin, with bright yellow eyes and intricate details in a surreal lounge setting.
Digital portrait of a young white man with white hair and blue eyes, dynamic and detailed new wave music album cover titled Counting Down The Days by MUSAIC.

photography Images

Grandpa with gray hair and woman with high ponytail gray hair standing in a turquoise pool with a white dog and a kangaroo near pink walls
Dark and moody modern architecture featuring tiered concrete structures illuminated by a warm orange glow at night.
A mystical nighttime landscape featuring ancient stone ruins with steps, a glowing full moon setting behind distant mountains, a winding river valley, glowing stars, and a warm, magical sky illuminated by cinematic lighting.
Photorealistic black Lincoln Continental with dark tinted windows on a wet New York City street during golden hour sunset, featuring glossy pavement and rainy atmosphere in film grain style.
A fashion photography portrait of a woman with platinum short hair, peeking from behind a dark doorframe, illuminated by cool blue cinematic lighting in a minimalist style.
Close-up double exposure portrait of a woman with mist and ocean waves blending with her profile in muted, cinematic colors
Close-up of a young woman with green eyes and freckles partially hidden behind a bush of vibrant purple flowers and green leaves, one flower covering one eye.
A photorealistic watercolor-style illustration of an African woman in white tunic and trousers wearing sunglasses, walking down stairs inside a massive black futuristic spaceship with cyberpunk and post-apocalyptic design elements under dramatic lighting.
Photorealistic 3D render of a figure standing inside a large, dark, cylindrical tunnel with futuristic technology elements and glowing orange lights, illuminated by cinematic and volumetric lighting effects.
Front-facing hyperrealistic photo of a dystopian android wearing a sculpted metal cyberpunk helmet with wires and cables, cloaked in black robes.

traditional art Images

A confident female Roman guardswoman wearing intricately detailed bronze battle armor, standing in a historic city street setting during daylight.
Oriental ink wash painting depicting dramatic waterfalls cascading through rugged mountain cliffs with misty atmosphere and faint calligraphy
A calico cat sitting serenely among vibrant Japanese watercolor style flowers under a large red sun with elegant ink wash effects.
Close-up of a girl with shiny hair and clear skin, eyes closed.
Anime girl with long blonde hair in a traditional Japanese school uniform with a blue bow
A white tiger and a girl in traditional clothing on a misty mountain background. AI generated image using stable diffusion.
A dramatic scene with a lone samurai standing on water, facing a traditional Japanese pagoda in the distance under a stormy sky with swirling clouds and mist, generated using Stable Diffusion.
AI generated image using Stable Diffusion of a beautiful woman dressed in traditional Chinese attire with flowing hair, detailed ancient clothing, and a serene oriental background with a pagoda and foliage.