Nodes/FiL_Design_ImageMind/🕵️ Optic Scanner
ComfyUI Node

🕵️ Optic Scanner

👁️ Optic Scanner — analyzes images with LLM vision models and generates optimized prompts for Z-Image, FLUX, SDXL, QWEN, Krea 2, and Ideogram 4.

By FiL-Design-Ai·Created 2 months ago·Updated 16 days ago· 3
🕵️ Optic Scanner
  • config
  • image
  • prompt
  • metadata_json
  • metadata_dict
agent⚪ None
agent_focus⚪ None
detail_levelnormal
languageen
model_typeAuto/None
video_duration0
video_aspectAuto
video_soundAuto
video_cameraAuto
prompt_modeAuto
photo_styleNone
nsfw_photo_styleNone
art_styleNone
nsfw_art_styleNone
seed-1
response_formattext
width0
height0
prompt
negative_prompt
custom_style
Category🎨 FiL Design/LLM

Inputs (23)

NameTypeDefaultDescription
configFIL_PROVIDER_CONFIGSettings received from the Loader node.
agentCOMBO⚪ NoneSubject domain — what is in the frame.
agent_focusCOMBO⚪ NoneCraft layer to weigh heavier on top of the agent: composition, light, fine detail, cinematic read, or the subject's state and movement.
detail_levelCOMBOnormalHow detailed the answer should be.
languageCOMBOenLanguage of the answer. Auto replies in the language of your prompt.
model_typeCOMBOAuto/NoneTarget generation model — adjusts prompt syntax, length, and format. Video is a universal profile for video models (MiniMax H2/H3, Wan, HunyuanVideo, LTX Video, Kling); MiniMax H3 gets a dedicated timeline shot-block profile.
video_durationINT00–20Requested clip length in seconds. 0 = Auto (the LLM decides). The range follows the model: MiniMax H3 accepts 4-15 whole seconds (API limit), universal Video 2-20. Shown only for video model types.
video_aspectCOMBOAutoAspect ratio written into the shot framing — for MiniMax H3 into the timeline header line ('10s, 16:9.'). Auto leaves the choice to the LLM. Shown only for video model types.
video_soundCOMBOAutoSound design mode. Auto = the LLM decides; Off = silent clip (no sound clause); Layered = mandatory ambience + foley + music clause for audio-capable models (H3, Veo). Shown only for video model types.
video_cameraCOMBOAutoPreferred camera move. The LLM builds the shot around it and may adapt per story stage — a preference, not a hard lock. Shown only for video model types.
prompt_modeCOMBOAutoAuto picks Hybrid or Two-Stage depending on whether a style is selected.
photo_styleCOMBONonePhotographic style overlay applied on top of the base description.
nsfw_photo_styleCOMBONoneAdult-only photographic style overlay (alternative to Photo style).
art_styleCOMBONoneArt style overlay applied on top of the base description.
nsfw_art_styleCOMBONoneAdult-only art style overlay (alternative to Art style).
seedINT-1-1–999999999999Provider-side generation seed, if supported. -1 lets the provider pick one.
response_formatCOMBOtextOutput shape: prose text, flat comma-separated tags, or a JSON object.
imageoptIMAGEImage or batch to analyze. Leave disconnected to generate from Prompt field.
widthoptINT00–16384Target image width in pixels. Helps tailor prompt composition to aspect ratio if > 0.
heightoptINT00–16384Target image height in pixels. Helps tailor prompt composition to aspect ratio if > 0.
promptoptSTRINGWITHOUT AN IMAGE — the idea itself, expanded into a finished prompt: ginger cat on a windowsill, rain outside abandoned metro station, morning light through a hole in the ceiling WITH AN IMAGE — what to focus on and what the text is for: emphasise the light and skin texture prompt for an album cover the cup on the table is the subject, not the person keep empty space on the right for a headline WON'T WORK (this comes from the image): replace the woman with a man make her stand instead of sit remove the second person Leave empty and the image is simply described in detail. Details: docs/scanner-prompts.md
negative_promptoptSTRINGRemoves WORDS from the generated text, not objects from the image. Pose, subject and composition cannot be changed here — use the prompt field. PUT HERE what the model adds on its own: cliches: masterpiece, best quality, 8k, highly detailed effects: bokeh, lens flare, dramatic vignette junk: watermark, logo, signature, text overlay defects: extra fingers, plastic skin, waxy texture judgements: beautiful, stunning, breathtaking Short nouns, comma-separated, 5-10 of them. Not sentences with 'no': under FLUX / Z-Image Turbo / Krea 2 / Ideogram 4 / Video the list is flipped into positive wording, and 'do not make it blurry' has nothing to flip — write blurry. Not sure what to put? Leave it empty, run once, and pick out the words that bother you in the answer. Details: docs/scanner-prompts.md
custom_styleoptSTRINGCustom style text. Overrides photo/art/nsfw preset styles.

Outputs (3)

NameTypeDescription
promptSTRINGGenerated prompt text.
metadata_jsonSTRINGJSON metadata string.
metadata_dictDICTParsed metadata dict.