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 glm-4.7-flash 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 glm-4.7-flash SDXL prompt generation ability is: 7.87/10

Prompts critiqued: 12
Average score: 7.87 / 10
Highest score: 8.5 / 10
Lowest score: 7.2 / 10

A critique of glm-4.7-flash SDXL prompt generations abilities:

glm-4.7-flash demonstrates a solid intermediate-to-strong aptitude for writing SDXL prompts, with a current session average of 7.87/10 across 12 evaluated prompts. The prompts consistently establish visually appealing core concepts, clear genre direction, and useful atmosphere cues, especially in cinematic lighting, fantasy/sci-fi mood, and material keywords like glass, moss, metal, wet asphalt, fur, and neon reflections. The strongest skill is creating coherent, immediately imageable scenes that SDXL can usually interpret well. However, glm-4.7-flash often relies on generic quality tags such as “8k,” “cinematic,” “highly detailed,” and “dramatic atmosphere” instead of more precise visual controls, and many prompts lack stronger camera/framing instructions, spatial hierarchy, and subject-specific design details. Reliability issues also recur around fragile requirements such as exact symbols, reflections inside glass, complex scale relationships, and abstract action phrases like “struggling” or “battling through.” Overall, glm-4.7-flash writes attractive and functional SDXL prompts, but would improve significantly by replacing broad aesthetic language with concrete composition, lens, pose, scale, texture, and background-priority details.

Notes for using glm-4.7-flash

• This model this model is fast I got on average 42 tokens per second on my AMD Strix Halo RyzenAI Workstation.

• I set the max tokens to 2500, 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.