This workflow requires my Comfyui-lmstudio-prompt-enhancer custom node which is installable via Comfy Manager.
🛈 IMPORTANT!!!
You will need to search for, download, and load the model ornith-1.0-9b-mtp in LM studio. You will also need to start the API server and generate an API Key. Learn how to configure LM Studio here.
My rating of ornith-1.0-9b-mtp SDXL prompt generation ability is: 8.39 /10
Prompts critiqued: 12
Average score: 8.39 / 10
Highest score: 8.8 / 10
Lowest score: 8.15 / 10
A critique of ornith-1.0-9b-mtp SDXL prompt generations abilities:
ornith-1.0-9b-mtp demonstrates strong aptitude for writing SDXL prompts, especially in producing clear, visually cinematic concepts with reliable subject-first structure. Across the 12 prompts, the author consistently shows good control of mood, lighting, atmosphere, materials, and genre language, with repeated strengths in photorealistic texture cues, dramatic color grading, volumetric effects, and coherent fantasy/sci-fi/cinematic framing. The prompts are generally efficient and SDXL-compatible, avoiding excessive clutter while giving enough visual direction to produce attractive outputs. The main weaknesses are recurring: some subjects remain underdescribed beyond broad labels like “futuristic spaceship” or “masterpiece,” composition is sometimes less precise than the lighting direction, and several prompts rely on fragile elements such as exact reflections, logos/emblems, or branded visual identities that SDXL may not reproduce faithfully. Overall, ornith-1.0-9b-mtp is a capable prompt writer with a strong eye for atmosphere and image appeal, currently strongest at cinematic mood-building and material description, but would improve further by adding more distinctive subject design, sharper focal hierarchy, and more robust camera/framing specificity.
Notes for using ornith-1.0-9b-mtp
This model this model is fast I got on average 44 tokens per second on my AMD Strix Halo RyzenAI Workstation. -generates detailed prompts, with minimal rambling. The generated prompts produces images with expert composition and color grading.
I set the max tokens to 5000, this higher value is to accommodate this model's reasoning.
If the image looks flat, check the logs in LM Studio, the last log entry will contain a JSON path x.choices[0].message.content if the content field is empty that means the LLM ran out of reasoning tokens, this can be resolved by increasing max_tokens.
this model tends to start the enhanced positive prompt's string with newline characters "\n" it does not appear to cause any problems.
