March 2025 Update

Hello friends! Quick update for you all. TLDR: The Latina RC/WG keyframes are taking longer than expected, but I’m still planning on releasing sometime in March. The attached image was generated with Chroma, a new furry stylized-art model that’s currently being trained and looks pretty good so far!

I’m still working on keyframes for the Latina RC. I realized that the last clothing transformation I did that was this detailed was for the French maid TG and that was 7 keyframes instead of 20! Keeping the lighting, background and character consistent while harmonizing the outfit from different angles is really proving to be a challenge. The feet are basically done now, though, and I’ve moved onto the body closeup shot. I can’t wait to see how Kling handles the transformation! I’m still planning on releasing in March since things should move pretty quickly once I’m done with the keyframes. Dialogue is minimal and I’m not going to train my own voice models this time.

I don’t usually do this, but there have been a lot of cool releases recently, so I wanted to update you on those. This isn’t meant to be a complete roundup, but more of a list of releases that I want to test at some point to see if they are usable for longer projects. If you guys want any help setting any of this stuff up, let me know here or on Discord! I would love to see your results!

• Hunyuan Video Keyframe Control Lora – this is an open-source Lora that allows a starting and ending keyframe to be used with the Hunyuan model. This is not an official Hunyuan release, so I don’t have very high hopes for it, but I’d still like to test it. Hopefully the official image-to-video model releases soon!

• Wan2.1 – This is another open-source, uncensored model that I want to test. I feel like it might not be quite as good as Hunyuan, but I’ll do some head-to-head comparisons at some point.

• Chroma – This is a furry, stylized-art model that’s currently being trained by Lodestone Rock, the main creator behind the FluffyRock family of models, my favorite non-realistic models. I’ve been messing with Chroma locally while I generate Latina RC keyframes in the cloud, and the attached image is my best result. It seems like it will be a really good model once it’s fully trained!

• Claude 3.7 Sonnet – This is a LLM (like ChatGPT) that released recently, and it’s the best model I’ve ever used for writing fiction!

• Grok 3 Voice Mode – I’m not that impressed with Grok 3 (another LLM) overall but having an uncensored voice mode like this is really interesting to me.

• Pika 2.2 – This looks like it allows starting and ending keyframes now so I need to add it to my video model test!

• GPT-4.5 – This is supposed to release to premium users next week. Early reviews are not great but I’d still like to try it.