PixelWave FLUX.1-schnell 04 - Apache 2.0!
Safetensor Files: 💾BF16 💾FP8 💾bnb FP4
GGUF Files: 💾Q8_0 🤗Q6_K 💾Q4_K_M
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 [email protected]. 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
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
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
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.
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 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.
Description
Over 30 LoRAs trained and added and block weight merged to improve quality of the output.
FAQ
Comments (21)
what loras are merged ?
They are LoRAs I trained. Since I haven't got fine tuning working yet, I divide up my training data into chunks, train a bunch of LoRAs and then merge into the model.
Yeah this information would really be helpful as when we go and add our own Loras after the fact to this model it GREATLY skews it.
This is the main difference between using merged models and fully fine tuned models to end users...for most people just using it out of the box its un noticeable but the moment we customize anything the quality falls off like a rock.
where is the gguf q8 schnell ?!
Uploaded now. I needed some beauty rest, I'm sure you would understand 🙂
keep it up bro!!! I have been jamming since your very first versions of this model and it never disappoints. thanks for your blood sweat and tears!
Using the NF4. First generation attempt is slow but once the models are loaded this model is the fastest I have used so far. Seems to provide useful results nearly every time.
I'm on an older system 32gb ram and a 4060 8gb gets me about 30-35 seconds per image.
Please give sample script to use with diffusers for your FLUX.1-schnell 01 model. Thank you.
Those 'diffusers' files are actually the gguf versions, since civitai does not yet support gguf files. You need to unzip the file. If you are using comfyui, place the file in the models/unet folder. Then you need the comfyui-gguf nodes and use the Unet Loader (GGUF) node to load the model.
Love this model so far. Can finally use my trained LoRA with 3.5 Flux Guidance without affecting image quality. It still has a bit of unnatural shadow around cheekbones sometimes (not often, just occasionally), but I guess that's just a problem of Dev that's rather hard to eliminate... :/
I downloaded the "Full Model nf4" but i found pixelwave_flux1_dev_Q4_K_M_01.gguf in it.
The 'diffusers' files are actually the Q8_0 and Q4_K_M GGUF versions. The site doesn't support uploading GGUF files yet.
Had perpetual issues with very bad hands all over the place using the Flux dev version. Shame as it's otherwise the sharpest flux model with great lighting awareness and great realism.
Thanks for the feedback. I will make sure the hands are better in the next version. The training tools are getting better and that makes it easier to train the model with less issues.
@humblemikey I noticed the model gets insanely sharp results as well with DEIS + Karras but at the same time it was also a big source of increasing the AI-errors like bad hands. Not sure if there's anything that can be done about it, but that particular setup, when it was working, would produce legitimately 10/10 cinematic gems for Flux standards.
So its a merge not a fine tune? Why is it marked under trained for filters?
It's a lot of LoRAs that I trained, merged, trained off that new model, etc. It's not other people's LoRAs.
@humblemikey No no i 100% get that (and love the merged model) but its still not a Checkpoint trained? Like you could make 1000 of your own LORAs and add them to Flux base but its still a merge.. Its just a little deceiving considering there is no "trained" Flux models. Just mashing the OG model + Loras.
I think the community understands that training the checkpoint via training multiple LoRAs and applying them to the model still counts as 'checkpoint trained'. The tag isn't 'fine tuned' after all. Speaking from experience, because I did both fine tune and LoRA training for SDXL, fine tuning is a lot easier than the LoRA method of training a checkpoint. Dump all the images in one folder and run. Don't have to worry about comparing different epochs, block weight merging to reduce errors, etc. I was so relieved once I got fine tuning working for SDXL because I didn't have to spend hours carefully tweaking LoRA recipes anymore.
please upload also on tensor.art
So far this is the only checkpoint that is making me want to use Flux. (The skin textures are realistic unlike all other checkpoints so far)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.