Info here is a work in progress.

This model is similar to fluffyrock-1088-megares-terminal-snr-vpred, in that it uses terminal SNR and v-prediction, but is trained on additional datasets of e6, laion, booru, gpt4mini. Due to this "diversity" of the very large datasets, the outputs can have a much wide range of styles and concepts than most models trained on just e6, though the results can also be somewhat "experimental" and unstable.

Unlike many other FR models, this one was not forked from another FR model and started over from SD1.5. Because of this and the increased size of the dataset, it takes a geometrically longer amount of time for a full training pass to complete and the model is much further behind (knowledge-wise) than other FR models.

All past epochs, and possibly newer, can be found at the HuggingFace Repo for E6-LAION

PolyFur is a somewhat similar project also being worked on, but with a more curated dataset.

Prompting:

The model understands natural language similar to SD1.5 and e621 tags, but as separate islands of knowledge, so using both may result in better output. Use of A1111 BREAK keyword recommended.

It may be necessary to prompt more completely (like not prompting for 'tail', what could have been a tail may end up as a 3rd leg) and carefully than with other models, even other FluffyRock models. E6LAION may interpret what is normally meaningless "prompt salad" literally and give (un)expected results.

Use e621 tags, without underscores, comma separated, any order.

Artist tags use "by name" format without "(artist)" on the tags that normally have them.

Note that additional setup is required to use FluffyRock vpred models:

Use the provided config file.

You will need to use cfg rescale.

For A1111 (and possibly the Vlad fork), use CFG_Rescale_webui extension.

There is a method to to do this on Comfy UI, but I haven't personally tested it yet. I believe ComfyUI_experiments has the node.

Autocomplete:

Tag Autocomplete File - this is currently only covering the pre-3M dataset. I am working on building a new one, but there's 35k conflicting tags I have to manually verify and correct.

Two Epoch Numbers?

First number is continuous from the start of training.

Second number is from when that specific line was forked.

For E6-LAION, the model forked from an earlier E6-LAION experiment with a smaller dataset.

Troubleshooting:

Output looks bad:

Do not sample at 512x512. Use 768 or greater. Going past 1088 may result in the typical SD1.x highres anomalies. High-res-fix and other similar methods work well to easily achieve 2k+ resolutions.

Prompt some art styles. Use some "by [e6 artist tag without underscores]". For better results, prompt several. Prompts without any form of prompted style may be terrible.

VPred troubleshooting:

Output just noise/cloud: Missing config file.

Output too dark: Turn up cfg rescale. Usually 0.7-0.9 seems to work best.

Some samplers may not work as cfg rescale support is not complete as of yet. See discord thread for latest discussion on this.

Training LoRAs:

vpred may require training with v_parameterization enabled. kohya_ss might complain about using that on v1, ignore that, nobody expected people would train SD1.5 with v-prediction.

LoRAs trained on non-vpred FR models will likely work.

Ask in the Discord for help.

Links and resources:

Tag Autocomplete File

Hugging Face Repo Contains every version of every model line. Full git clone of the repo will require >1.5TB disk space, you have been warned.

FluffyRock Discord Server

Furry Diffusion Discord Server and the FR thread there

LodestoneRock's Patreon Help support them with the cost of training.

License: WTFPL

Due to Civitai's on-site gen being broken (for at least these models), I've had to set the commercial use to an incorrect value to disable the annoying "Create" button. You can use the models on gen service, we don't care, but it would be cool if it actually works. :V

Apparently it works now except with vpred models.

If you release a public merge of FluffyRock models, a link back to the HF and/or Civitai post for FR would be cool.