ComfyUI Prompt Composer adds RefMod and LoRA support for subject-based prompt orchestration
Francky_B · reddit · 2026-09-24
A developer has shipped a major update to the Prompt Composer node in the open-source ComfyUI-Prompt-Manager plugin, organizing prompts as reusable snippets bound to subjects:
- RefMod & LoRA support: each snippet can carry one image LoRA and one video LoRA, with the node outputting the right ones via an Image/Video toggle; includes a visual RefMod picker, batch RefMod generator, and subjectdefinitions / custom prompthint output for Minimax.
- Subject grouping: characters and their clothing/hair/expression snippets stay grouped as <Subject 1>, <Subject 2>; snippets can be drag-and-dropped between subjects; in image mode subject prompts come first, followed by actions, lighting, and compositing.
- Random prompts: multiple prompts can be bundled into one 'random' slot, with LoRAs/RefMods switching automatically.
- Thumbnails: the Prompt Browser generates prompt thumbnails (using image LoRAs where set); sample prompts were made with Krea 2 Turbo.
Works with most models including Krea 2 (not yet Anima or Qwen); for Minimax the author recommends limiting it to subjects and feeding detaileddescription as separate text. Ships with presets and an example workflow on GitHub.
More from coding & agent
- The whole voice agent demo cost ~$0.02: KugelAudio at $0.035/min, Gladia $0.75/hr — tobowers · 2026-09-24
- A real-time voice agent with zero US servers: Gladia STT, Gemma 4 on Scaleway, KugelAudio TTS — tobowers · 2026-09-24
- Devs add 10x more test harnesses just to slow down AI coding agents — cjimti · 2026-09-24
- Screenshot-annotated web edits tested with Ling-3.0: only 2 of 3 runs passed — alifcoder · 2026-09-24
- Claude directory approved an MCP submission in just 5 minutes — devenbhooshan · 2026-09-24
- Usage resets land before rest days, forcing devs into all-night AI coding sessions — cjimti · 2026-09-24