How It Works

• You describe what you want in natural language
• The plugin calls the HuggingFace Inference API with your prompt
• The image is saved locally as a timestamped .png or .jpeg file
• The file path is returned so LM Studio can display it

Free tier note: HuggingFace free accounts can use FLUX.1-schnell and SDXL. Cold-start on an inactive model may take 20–60 seconds on the first call. Subsequent calls are fast.

Pro tier note: FLUX.1-dev and FLUX.2-dev require HuggingFace Pro credits. They produce noticeably higher quality output but cost credits per generation.

Tools

generate_image — Generate from text

Creates an image from a text prompt and saves it to disk.

generate_image(prompt, model_id?, negative_prompt?, lora_id?, lora_scale?)

Returns:

• file_path — absolute path of the saved image
• model_used — which model was used
• lora_used — LoRA adapter applied, if any
• file_size_bytes, mime_type

LoRA note: LoRA generation routes through the fal-ai provider, which supports FLUX LoRA compositing. Pair FLUX LoRAs with a FLUX base model.

list_models — Show available models

Returns the curated list of popular text-to-image models with speed, style, and access tier.

list_models()

Use when the user asks what models are available or wants to pick a different one.

Available models:

You can also pass any HuggingFace text-to-image model ID directly to generate_image without it appearing in this list.

list_loras — Search LoRA adapters

Searches HuggingFace for LoRA adapters compatible with FLUX base models.

list_loras(search?)

Returns a list of LoRA model IDs sorted by downloads. Pass the id field as lora_id in generate_image.

Example Prompts

Photorealistic portrait:

"A cinematic close-up portrait of a woman in golden hour light, soft bokeh background, 85mm lens, film grain"

Concept art:

"A futuristic megacity at night, neon reflections in rain puddles, cyberpunk aesthetic, ultra-detailed, 4K"

With negative prompt:

prompt: "A peaceful mountain lake at sunrise" negative_prompt: "people, cars, buildings, text, watermark"

With LoRA style:

First: list_loras("anime") → find a LoRA ID Then: generate_image(prompt="...", lora_id="alvdansen/flux-koda", lora_scale=0.8)

Using a specific model:

generate_image(prompt="...", model_id="stabilityai/stable-diffusion-xl-base-1.0")

Prompt Tips

• Be specific — vague prompts produce average results. Name the subject, lighting, style, mood.
• Quality terms work: ultra-detailed, 8K, cinematic, professional photography, sharp focus
• Style references help: in the style of Studio Ghibli, watercolor illustration, photorealistic, oil painting
• Negative prompts clean up common issues: blurry, low quality, deformed hands, text, watermark, oversaturated
• FLUX models understand natural language well — you can write full sentences rather than comma-separated tags

Accepting Model Licenses

Some models require accepting their license on HuggingFace before they can be used:

• Go to the model's page on huggingface.co (e.g. huggingface.co/black-forest-labs/FLUX.1-schnell)
• Click "Access repository" and accept the license
• Make sure your API token has at least read scope

This only needs to be done once per model per account.