[New Model]: circlestone-labs/Anima

The model to consider.

https://huggingface.co/circlestone-labs/Anima

Anima is a 2B text-to-image diffusion model for anime, character, and illustration generation.

The closest model vllm-omni already supports.

Closest model: Qwen/Qwen-Image, supported by QwenImagePipeline.

Similarity: Anima is also a text-to-image diffusion model, and its released files include qwen_image_vae.safetensors, so part of the Qwen-Image VAE path may be reusable.

Difference: Anima is is packaged as a Diffusion Single File model, while Qwen/Qwen-Image is supported through vLLM-Omni's native Qwen-Image pipeline/repo layout.

What's your difficulty of supporting the model you want?

The main difficulty is Diffusion Single File support. vLLM-Omni currently supports repo-style diffusers loading through DiffusionPipeline.from_pretrained(), but does not support Diffusion Single File checkpoints directly.

Likely patch points:

• vllm_omni/diffusion/models/diffusers_adapter/pipeline_diffusers_adapter.py: DiffusersAdapterPipeline.load_weights() currently always calls DiffusionPipeline.from_pretrained(model_id, **load_kwargs). It needs a single-file path that calls DiffusionPipeline.from_single_file(...).
• vllm_omni/diffusion/data.py: OmniDiffusionConfig.enrich_config() currently expects diffusers repo metadata like model_index.json. Single-file support needs a way to skip or override that metadata path and provide the pipeline class/config explicitly.

Use case and motivation

Anima is a popular anime / illustration-focused text-to-image model. Supporting it would expand vLLM-Omni's image generation coverage and allow serving Anima through the existing image generation or diffusion chat APIs.

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.