cant add Anima model

[lqp]

I am not sure whether this is configuration error on my side or bug in the code, so I post here.

I can't make Anima diffusion model (https://huggingface.co/circlestone-labs/Anima) to work under LocalAI. Neither under diffusers backend, nor under stablediffusion-ggml backend.

Anima worked under pure stalbediffusion.cpp using such commandline (from docs):

./sd-cli --diffusion-model anima-preview.safetensors --vae qwen_image_vae.safetensors --llm qwen_3_06b_base.safetensors -p "a lovely cat holding a sign says 'anisnuff.cpp'" --cfg-scale 6.0 --sampling-method euler -v --offload-to-cpu --diffusion-fa

But when I trying to run it from LocalAI using this config.yaml

---

backend: stablediffusion-ggml
name: anima-cpp
cfg-scale: 6.0
options:

• diffusion-model
• vae_path:anima/qwen_image_vae.safetensors
• llm_path:anima/qwen_3_06b_base.safetensors
• lora_dir:anima/l/
• sampling-method:euler
• diffusion-fa
  parameters:
  model: anma/anima-preview.safetensors

---

it gives me a long list of errors like this: "[ERROR] model_manager.cpp:640 - Diffusion model tensor 'model.diffusion_model.net.t_embedder.1.linear_2.weight' not in model metadata".

When I trying to run it with diffusers this diffusers config.yaml (diffusers should load additional files, right?):

---

backend: diffusers
known_usecases:
- Image
diffusers:
cuda: true
enable_parameters: negative_prompt,num_inference_steps,clip_skip
pipeline_type: AnimaModularPipeline
scheduler_type: k_dpmpp_2m
f16: true
name: anima-diff
parameters:
model: anima/anima_preview.safetensors
step: 25

I get errors: "ERROR artifact materialization failed; the backend will download the model in-band on every load model="anima-diff" backend="diffusers" error=resolve Hugging Face revision: Hub request returned status 401".

[lqp]

if I add to stablediffusion-ggml config option

• diffusion_model_path:anima/anima-preview.safetensors

then error go away and it begin to generate, but generates something senseless, even remotely not following the prompt and not a meaningful images but some blobs.