CHIMERA

Chimera is an SDXL anime model merge that supports Danbooru-style artist tags. It doesn't require the use of meta tags (e.g. score_6, masterpiece, very aesthetic) in the positive prompt to get good results. These are optional (except for score_X pony tags, these are not active). For negative prompt, the following should be used at the very least.

worst quality, low quality, bad anatomy, watermark, signature

Artist styles are recommended to get non-generic-looking results.

Merged models:

1. CashMoney (Anime) v.1.0

2. Pony Diffusion V6 XL

3. Animagine XL V3.1 (and v3.0)

4. Anime Illust Diffusion XL

5. ArtiWaifu Diffusion - v1.0

6. Godiva - v2.0

7. 0003 - Pony - 0003-delta

Features:

• Amplified support for artist styles. See example images for examples. It is recommended you first use (artist name:0.5) or (by artist name:0.5) and adjust as necessary.

• No strict need for meta-tags (e.g. score_6, masterpiece, very aesthetic). Do not use pony score_X tags.

• Improved anatomy over base merged models using realistic model Godiva.

• Optimal CFG scale optimised at 9 (7 to 10 recommended).

• Generates in semi-realistic styles as well as traditional styles. Use combinations of realistic, 2d, 3d, etc tags in positive or negative prompt for effect.

• Artist style mixing is highly effective at producing unique and original-looking results.

Known issues:

• Some artist styles aren’t fully accurate.

• Pony-based LoRAs may not function as intended, especially with more complex prompts. Some work, others don’t. SDXL-based LoRAs work however. Note: If the Pony LoRA has been exclusively or substantially trained on OUT01-05 it will not work.

• The model is less adept at complex NSFW than Pony Diffusion.

• Limited furry-gen support and no pony-gen support.

License:

FAIPL 1.0

Merge strategy:

Coming soon.

Credits:

Thank you to all of the model creators and teams which produced the high-quality models for this merge.

Special thanks to sulph.