Anima-2.9B Released: A Fine-Tuned Model for Anime and Illustration
RevolutionaryWater31 · reddit · 2026-08-12
Anima-2.9B is a fine-tuned and layer-expansion model based on Anima, targeting anime, illustration, and non-photorealistic art. It continues training on the original architecture with an additional 1.7M anime/illustration samples, with a knowledge cutoff in July 2026.
Training Details:
- Trained using the Muon optimizer on an 8x 5080s cluster.
- In Preview v1, only new layers were trained, with 70% of compute spent on 1024px resolution.
- Utilized mixed captioning (tags and natural language) assisted by Gemini 3.1/3.5 Flash-Lite and Claude Sonnet 5.
The model is available on HuggingFace with plug-and-play support for ComfyUI custom nodes.
More from Multimodal
- KREA2 LoRA: Generate Instagram Influencer Aesthetic Photos Out of the Box — vizsumit · 2026-08-12
- Real-Time Hand Gesture Control for AI Video Effects via MediaPipe and Gemini — mhdfaran · 2026-08-12
- MiniMax H3 Stuns in Img2Vid Test: Generating Batman Action From a Single Image — OohFekm · 2026-08-12
- MiniMax H3 Video Model Accused of Frequent Unwanted Cuts Despite Explicit Prompts — johnk419 · 2026-08-12
- Lightricks Releases LTX-2: Open-Source Audio-Video Generative Model and LoRA Trainer — Lightricks · 2026-08-12
- Looking for LTX 2.5 IA2V Workflow — Pitiful_Owl9068 · 2026-08-12