Image & Video

Anima Base Model surprises with non-anime versatility, LoRA training on 12GB GPUs

User trains first LoRA on 12GB 3060, generates mixed Baroque-90s 3D styles

Deep Dive

A Reddit user, skeptical of Anima's anime focus, decided to test its versatility for non-anime art. Initially, artist tags like '@moebius' failed, so they trained their first LoRA on a 12GB RTX 3060 using Anima Trainflow. The process was fast and easy, yielding strong results. They then explored prompting by mixing styles like Baroque painting with 1990s 3D games, generating 20-40 images per seed. The variety in styles and faces per seed was impressive.

To refine prompts, the user uploaded favorite images to Gemini (or ChatGPT for better analysis) for detailed style descriptions, then fed those back into Anima. This iterative approach produced consistent, high-quality outputs. The user confirms Anima's claim of extreme style versatility—it only needs a slight nudge to adopt almost any artistic style, even without anime. The LoRA training example will be shared separately.

Key Points
  • Trained first LoRA on 12GB RTX 3060 using Anima Trainflow
  • Generated mixed styles like Baroque + 1990s 3D games per seed
  • Used Gemini/ChatGPT to convert output images into detailed style prompts

Why It Matters

Shows Anima's non-anime potential for artists with modest GPUs, enabling custom style fine-tuning.