Verboa Prompt Writer: a small language model that writes prompts for Verboa Image 1.0. Tell it as much or as little as you like (who, how old, where, how explicit, voice, length) and it writes the prompt the way a person would type it. It runs inside ComfyUI, right in front of the image model, and writes a fresh prompt every run.<h2 id="three-modes">Three modes</h2><ul><li>Let Me Choose: nine fields (people, age, place, level, ethnicity, body, camera, voice, length). Leave any on "any" and the writer invents it.</li><li>Surprise Me: no fields; every run spins them at random.</li><li>Write My Own: your own prompt goes through the same workflow.</li></ul><h2 id="setup">Setup</h2><ol><li>Put the model file in ComfyUI/models/text_encoders/: fp8 (3.1 GB, NVIDIA) or bf16 (4.4 GB, Mac or any GPU).</li><li>Install the node and workflows from the Verboa Image - ComfyUI Workflows page (civitai.red/models/2981936), version v1.2 "Prompt Writer node + writer workflows (zip)": copy the ComfyUI-Verboa folder into ComfyUI/custom_nodes/ and restart (needs ComfyUI 0.19.0 or newer).</li><li>Load the writer workflow (NVIDIA or Mac), pick a mode, press Run.</li></ol><h2 id="details">Details</h2><ul><li>Full fine-tune of Qwen3.5-2B-Base on 35,206 of the captions Verboa Image 1.0 was trained on. About 18 tokens per second on a laptop RTX 5090 (1.5 to 16 seconds per prompt).</li><li>Sampling: temperature 1.0, min_p 0.05, repetition penalty 1.05.</li><li>For the image model itself see the Verboa Image page: 8 steps, CFG 2, shift 4.0.</li></ul>18+ only. Fictional adults only.

Verboa Prompt Writer - NSFW prompt generator for Verboa Image (ComfyUI)

fp8 file for NVIDIA cards (RTX 40/50: fp8 is native). Put it in ComfyUI/models/text_encoders/ and load it with a Load CLIP node (type stable_diffusion). Needs the ComfyUI-Verboa node (see the node + workflows version).

Files with SHA256: d93a51bc855f9d99b0a8edd1c4c48b632dc6883906699bfb461691aa393f2b75