Foreign-Mirrors/Fooocus
1
0
Fork 0
mirror of https://github.com/lllyasviel/Fooocus.git synced 2026-08-16 13:13:16 +02:00
Code Issues Packages Projects Releases Wiki Activity

Fooocus Prompt Expansion (#329)

Browse Source 
* add vae approx download

* files

* files

* files

* i

* i

* i

* i

* i

* i

* i

* i

* i

* i
This commit is contained in:
lllyasviel
2023-09-09 17:22:32 -07:00
committed by GitHub
parent 09e0d1cb3a
commit 496766edd7
19 changed files with 150532 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files
readme.md
+10 -11
View File
@@ -97,17 +97,16 @@ Coming soon ...
Below things are already inside the software, and **users do not need to do anything about these**.
~Note that some of these tricks are currently (2023 Aug 11) impossible to reproduce in Automatic1111's interface or ComfyUI's node system.~ (Update Aug 21: We are working on implementing some of these as webui extensions/features.)
1. Native refiner swap inside one single k-sampler. The advantage is that now the refiner model can reuse the base model's momentum (or ODE's history parameters) collected from k-sampling to achieve more coherent sampling. In Automatic1111's high-res fix and ComfyUI's node system, the base model and refiner use two independent k-samplers, which means the momentum is largely wasted, and the sampling continuity is broken. Fooocus uses its own advanced k-diffusion sampling that ensures seamless, native, and continuous swap in a refiner setup. (Update Aug 13: Actually I discussed this with Automatic1111 several days ago and it seems that the “native refiner swap inside one single k-sampler” is [merged]( https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12371) into the dev branch of webui. Great!)
2. Negative ADM guidance. Because the highest resolution level of XL Base does not have cross attentions, the positive and negative signals for XL's highest resolution level cannot receive enough contrasts during the CFG sampling, causing the results look a bit plastic or overly smooth in certain cases. Fortunately, since the XL's highest resolution level is still conditioned on image aspect ratios (ADM), we can modify the adm on the positive/negative side to compensate for the lack of CFG contrast in the highest resolution level. (Update Aug 16, the IOS App [Drawing Things](https://apps.apple.com/us/app/draw-things-ai-generation/id6444050820) will support Negative ADM Guidance. Great!)
3. We implemented a carefully tuned variation of the Section 5.1 of ["Improving Sample Quality of Diffusion Models Using Self-Attention Guidance"](https://arxiv.org/pdf/2210.00939.pdf). The weight is set to very low, but this is Fooocus's final guarantee to make sure that the XL will never yield overly smooth or plastic appearance (examples [here](https://github.com/lllyasviel/Fooocus/discussions/117)). This can almostly eliminate all cases that XL still occasionally produce overly smooth results even with negative ADM guidance. (Update 2023 Aug 18, the Gaussian kernel of SAG is changed to an anisotropic kernel for better structure preservation and fewer artifacts.)
4. We modified the style templates a bit and added the "cinematic-default".
5. We tested the "sd_xl_offset_example-lora_1.0.safetensors" and it seems that when the lora weight is below 0.5, the results are always better than XL without lora.
6. The parameters of samplers are carefully tuned.
7. Because XL uses positional encoding for generation resolution, images generated by several fixed resolutions look a bit better than that from arbitrary resolutions (because the positional encoding is not very good at handling int numbers that are unseen during training). This suggests that the resolutions in UI may be hard coded for best results.
8. Separated prompts for two different text encoders seem unnecessary. Separated prompts for base model and refiner may work but the effects are random, and we refrain from implement this.
9. DPM family seems well-suited for XL, since XL sometimes generates overly smooth texture but DPM family sometimes generate overly dense detail in texture. Their joint effect looks neutral and appealing to human perception.
1. Prompt expansion and the "Raw mode" to turn it off (similar to Midjourney's hidden pre-processsing and "raw" mode, or the LeonardoAI's Prompt Magic).
2. Native refiner swap inside one single k-sampler. The advantage is that now the refiner model can reuse the base model's momentum (or ODE's history parameters) collected from k-sampling to achieve more coherent sampling. In Automatic1111's high-res fix and ComfyUI's node system, the base model and refiner use two independent k-samplers, which means the momentum is largely wasted, and the sampling continuity is broken. Fooocus uses its own advanced k-diffusion sampling that ensures seamless, native, and continuous swap in a refiner setup. (Update Aug 13: Actually I discussed this with Automatic1111 several days ago and it seems that the “native refiner swap inside one single k-sampler” is [merged]( https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12371) into the dev branch of webui. Great!)
3. Negative ADM guidance. Because the highest resolution level of XL Base does not have cross attentions, the positive and negative signals for XL's highest resolution level cannot receive enough contrasts during the CFG sampling, causing the results look a bit plastic or overly smooth in certain cases. Fortunately, since the XL's highest resolution level is still conditioned on image aspect ratios (ADM), we can modify the adm on the positive/negative side to compensate for the lack of CFG contrast in the highest resolution level. (Update Aug 16, the IOS App [Drawing Things](https://apps.apple.com/us/app/draw-things-ai-generation/id6444050820) will support Negative ADM Guidance. Great!)
4. We implemented a carefully tuned variation of the Section 5.1 of ["Improving Sample Quality of Diffusion Models Using Self-Attention Guidance"](https://arxiv.org/pdf/2210.00939.pdf). The weight is set to very low, but this is Fooocus's final guarantee to make sure that the XL will never yield overly smooth or plastic appearance (examples [here](https://github.com/lllyasviel/Fooocus/discussions/117)). This can almostly eliminate all cases that XL still occasionally produce overly smooth results even with negative ADM guidance. (Update 2023 Aug 18, the Gaussian kernel of SAG is changed to an anisotropic kernel for better structure preservation and fewer artifacts.)
5. We modified the style templates a bit and added the "cinematic-default".
6. We tested the "sd_xl_offset_example-lora_1.0.safetensors" and it seems that when the lora weight is below 0.5, the results are always better than XL without lora.
7. The parameters of samplers are carefully tuned.
8. Because XL uses positional encoding for generation resolution, images generated by several fixed resolutions look a bit better than that from arbitrary resolutions (because the positional encoding is not very good at handling int numbers that are unseen during training). This suggests that the resolutions in UI may be hard coded for best results.
9. Separated prompts for two different text encoders seem unnecessary. Separated prompts for base model and refiner may work but the effects are random, and we refrain from implement this.
10. DPM family seems well-suited for XL, since XL sometimes generates overly smooth texture but DPM family sometimes generate overly dense detail in texture. Their joint effect looks neutral and appealing to human perception.
## Advanced Features