v9.28
🔑 API keys — checking…tap to edit
Get one at aistudio.google.com/apikey · used for NBP/NB2 image generation.
Used only for forensic target analysis when an OpenAI analyzer is selected.
Used only for the Seedream 5.0 Pro body pass. All keys stay only in this browser's local storage.
1Target & analysis
About this pipeline
Model-only swap: the target photo and selected Apoorva references go straight to the selected Google image model. The pose & realism spec below is optional — one cheap Flash-Lite call per target.
Tap to upload the photo to work on
Send resolution
Renderer read
About send resolution
Only the temporary API copy changes. Your stored original stays untouched.
Optional — pose spec (classic)tap to open
About pose spec
On = one cheap Gemini Flash-Lite call (~$0.002, cached per target photo) reads the target and writes pose mechanics + the photo's capture signature as text into the prompt — so the renderer is never asked to study the target's face. Off = classic behaves exactly as before.
Identity lockadvanced · forensic only
About identity lock
On = her fixed line is pinned into the prompt, replacing the target's trait. Turn on only what this shot actually shows (skip nails for close-ups, etc.). Off = the scene and refs decide. The pinned short line goes into the prompt header; the rewriter replaces the target's trait in the scene text. Age words (young, teen, girl…) are always scrubbed — they pass the prompt check but get the finished image confiscated on the way out.
Edit identity lines
About identity backup
Backup saves all seven identity lines, header short-lines and toggle states to one .json — restore it after app updates or on another device and your Apoorva database comes back exactly as it was.
2Character references for this run
Send resolution
Renderer read
Send resolution
Renderer read
Send resolution
Renderer read
About reference picks
Pick the refs that match this shot — e.g. smiling face for a smiling target, back body for a back-view target. The prompt assembles itself from your picks.
View assembled prompt (advanced — edits apply to this run only)
3Generate
About models
NBP remains the safest identity/fidelity choice. NB2 is the faster Google renderer and supports the same multi-reference pipeline. GPT Image 2 runs on your OpenAI key — classic mode sends face, body and target in that order with its own fixed prompt (editable in the prompt box); no temperature applies.
Advancedretry
About retries
Auto-retry re-runs a failed or confiscated render up to 3 times with a fresh draw — covers safety confiscations and iOS killing the call when you switch tabs or lock the screen. Every attempt is billed separately.
Session cost
4Result
Seedream 5.0 Pro body pass
About Seedream body pass
Manual only. Image1 is the image currently shown above. Image2 is the currently selected Classic body reference, sent from the original image stored in the app. The 1K/2K setting controls only Seedream's output resolution.
About carousel anchor
Anchors this image as the keeper: carousel variations take identity, body, outfit and accessories from it as pixels, and only what changes from each new target.
📚 Reference librarystored on this device · reused every run
About the library
Upload once — stored in the app's database on this device, reused every run. Faces: one photo per expression (neutral, smiling…). Bodies: one per view (front, side, back…). Face/body captions use their existing prompt paths; Phone uses its fixed prompt line.
🕘 Historyevery run saved automatically
No runs yet — every generation is saved here automatically, with its prompt, settings and refs, until you delete it.
About analysis export
Export produces one .zip: all result images + a manifest with the exact prompt, settings, refs and your notes for every run. Upload that zip in a new chat and it can be analysed — what worked, what failed, patterns across runs.