Inspecting and Preparing Images

Campfire includes local image-analysis and image-utility capabilities that help CodeBot compare images, inspect generated assets, prepare masks, crop references, build contact sheets, and check whether a generated character still looks like the intended person.

These run locally inside Campfire and return compact results such as scores, verdicts, face boxes, masks, crops, and metadata summaries.

Ask CodeBot to Check Image Quality or Duplicates

CodeBot can compare perceptual hashes, report simple blur or exposure signals, compare numeric vectors, and check a group of images for likely duplicates or hygiene problems.

Typical prompts:

• "Check these generated images for duplicates."
• "Which variants are blurry or overexposed?"
• "Tell me whether these two images are basically the same."

Ask CodeBot to Make Contact Sheets, Crops, or Grid Splits

CodeBot can create a labeled contact sheet, crop one or more regions from an image, or split a generated grid into separate files.

Typical prompts:

• "Make a contact sheet from these variants."
• "Crop each face into a separate reference image."
• "Split this 2x2 generated grid into individual files."

These workflows are useful before choosing a final image or preparing references for another generation step.

Ask CodeBot to Inspect or Strip Metadata

CodeBot can inspect image metadata, including likely AI-generation metadata such as prompt, workflow, seed, model, sampler, or steps when present. It can also strip shareable container metadata.

Typical prompts:

• "Find the prompt and seed used for this image."
• "Remove metadata before I share this file."
• "Check whether this image contains workflow data."

Metadata stripping removes ordinary file metadata exposed through the image container. It does not guarantee removal of invisible pixel-level watermarks or visual signatures.

Ask CodeBot to Find Faces and Create Masks

CodeBot can detect faces, show a visual picker so you can choose a face, and create masks for face repair or selected image regions. It can also create clothing-like masks for portrait-style edits.

Typical prompts:

• "Let me pick which face to repair."
• "Create a mask for the eyes."
• "Make a mask for the shirt in this image."

Use these before image-editing or inpainting workflows that should affect only part of an image.

Ask CodeBot to Check Character Identity

CodeBot can compare face identity using local ArcFace models. This helps answer whether a generated scene still contains the intended character.

Typical prompts:

• "Does this generated face match the reference character?"
• "Find which faces in this group match Lina."
• "Compare these character sheets and tell me what needs review."

Use face identity matching for character continuity. Do not use broad visual similarity as proof that a face is the same person.

Ask CodeBot to Compare Style or Composition

CodeBot can compare whole-image similarity using CLIP image embeddings. This is useful for style, composition, or semantic similarity.

Typical prompts:

• "Which variant is closest to this reference style?"
• "Compare these images for overall similarity."
• "Pick the image that best matches this composition."