How to Load FLUX 1‑dev in LM Studio

1. Install / update LM Studio

   • Download the latest LM Studio release (≥ 0.3.14) from the official site and launch it.
   • Newer builds include built‑in GGUF support and the Local Server mode.

2. Enable the Local Server

   • Open the Developer tab → toggle “Local Server” on.
   • This exposes an OpenAI‑compatible endpoint (e.g., http://localhost:1234/v1).

3. Obtain a FLUX 1‑dev checkpoint

   • Preferred: FLUX 1‑dev GGUF Q8 (≈ 24 GB) – see the guide on quantized GGUF loading ⁷.
   • Alternative: the prompt‑crafter‑for‑flux.1‑dev repository which bundles ready‑to‑use prompts ¹.

4. Add the model to LM Studio

   • In the Models pane click “Add Model”, browse to the downloaded flux1-dev-q8.gguf file, and give it a name (e.g., “FLUX‑dev”).
   • LM Studio will index the model and list it for selection.

5. Configure model settings

   • Increase max tokens (e.g., 1024) so the image payload isn’t truncated.
   • Enable vision support if you plan to use text‑to‑image or in‑painting features.

6. Test the model

   • In the chat UI, send a prompt such as:

     A futuristic cityscape at sunset, photorealistic, 8K
     

   • LM Studio returns a base‑64‑encoded PNG (or a file path) that can be viewed directly.

7. Optional: Call the model from code

   • Use any HTTP client to POST to http://localhost:1234/v1 with JSON:

     {
       "model": "FLUX‑dev",
       "prompt": "A cyberpunk street market, neon lights"
     }
     

   • The response contains the image data; decode or save it as needed. See the tool‑use docs for details ⁶.

Common Pitfalls & Fixes

• Memory constraints – The full model needs > 24 GB VRAM; use the Q8 quantized file or run on CPU (slower).
• Version mismatch – Older LM Studio versions cannot load GGUF files; updating resolves the issue.
• Model not appearing – Ensure the file extension is .gguf and the folder is not nested; LM Studio only scans the top‑level directory you select.