Pony V7 - Pony Image Samples [WIP, still adding images]

Examples of Pony V7 RC0 handling Pony specific prompts.

The model is still being trained but after many months of experimentation we are finally getting closer to the release so I will be posting examples of V7 generation grouped by themes:

• Pony [you are here]

• Furry [coming next]

• Anime

• Realism

• Backgrounds

• Interesting prompts / spatial prompts / style clusters

V7 prompting is much more precise than V6 (or most pre T5 models) so these samples are using a combination of captions produced from images and LLM generated to simplify the prompting, you don't need to prompts like this and we will be shipping a finetuned LLM to do this automatically but it's worth demonstrating the high end of prompting capabilities.

Some of the Pony highlights you can see here:

• 1.5k image resolution

• Much stronger prompt understanding

• Style prompting

• Ability to generate very light or very dark scenes

• Better small details and backgrounds (or lack of them when not asked for one)

• Better hands and complex anatomy (upside down characters)

• Generalization of characters (i.e. you can describe someone instead of using a name of specific known character)

All images are generated using 8bit GGUF version of the model on a 4090 using 40 steps at around 80-90s/image.

Please join us at http://purplesmart.ai/discord to be the first to try V7