Design API CALL Panel

API CALL Panel

API CALL is where the user chooses a model and sets its inputs. The name says what it builds: the exact request the next Generate will send. Opening the panel costs nothing. Only the Suggest setup tool sends anything, and it asks first.

The panel fills the screen, because a model can have dozens of inputs. It remembers its scroll position while it redraws, so checking a box never throws the user back to the top. It also keeps the keyboard closed until the user taps a field.

Top bar and banners

Box	What it holds	Why
Close (X)	Leaves without saving	A plain exit for browsing. Unsaved changes are dropped on purpose, because Apply is the only commit.
Title and subtitle	"API CALL" and "Choose the model and inputs for your next request."	States the panel's single job
Progress bar	Shows while inputs load, a photo uploads, or a suggestion runs	Explains why the controls are grayed out
Notice banner	A blue card with the latest result, such as "Inputs refreshed" or "Photo uploaded and set as image"	Sits above the scroll area, so the user sees the outcome without scrolling. Screen readers announce it.

Create card

Box	What it holds	Why
Chat, Image, Video	A segmented button, words only	Lets the user set up another mode without closing the panel. Unapplied edits in each mode are kept until Apply or Close.

Model card

Box	What it holds	Why
Base URL	The provider address under the card title, such as https://api.replicate.com/v1	Confirms which provider and which key the request will use
Find a model by features	A button that opens two rows of filter chips (Image and Video only)	Answers "which model can do this?" without reading every model page
Resolution chips	Any, 480p, 720p, 1080p	Video price and quality depend on resolution, so it is the first filter
Feature chips	Editing, Starting frame, Ending frame, Frame rate, Fast generation, LoRA, Image guidance, Audio input, Negative prompt, Seed, Safety controls	Each chip keeps only models whose real inputs support that feature. Editing keeps the nine editing models, for manual control of an edit.
Saved models	A dropdown of the preloaded models plus any the user added. It becomes "Matching models (n)" while filters are on.	One list for every model on this provider
Mismatch warning	A red line when the current model fails a selected filter	Stops a user from applying a model that cannot do what they asked for
Add a model	A field for any owner/model ID, with an Add button	Any Replicate model with a compatible input schema can join the list. The app remembers it.
WAN 2.1 resolution	A dropdown that appears only for the WAN 2.1 models	On WAN 2.1, 480p and 720p are separate models. Choosing a resolution switches the model, which is how that provider works.

LoRA card

This card appears only for black-forest-labs/flux-dev-lora.

Box	What it holds	Why
Saved FLUX dev LoRAs	A numbered dropdown of LoRAs from the LoRA Library, with each trigger word	Fills lora_weights and lora_scale in one step. The numbers keep two LoRAs with the same name apart. A notice then reminds the user to put the trigger word in the prompt.

Input cards

The panel opens in a simple view. It shows the model by a plain name and price, such as "Quick draft (FLUX schnell), under 1 cent", with its provider ID in small type. Below come only the settings most people change: Shape, Number of pictures, Length and Resolution for video, the Safety checker, picture inputs, and any required input. A picture input shows "A picture is attached" with Remove the picture, not a web address. Show all settings opens every input, the model finder, Add a model, and the Tools card; Show fewer settings goes back, and the app remembers the choice. Settings hidden by the simple view keep their saved values when the user taps Apply.

Some settings count only together with another. The panel reads this from the model's own description of a setting ("only used when size='custom'"), or, for Width and Height, from a list beside them that offers "custom". For Seedream 4.5, Width and Height appear only when Size is custom, and Shape only when it is not; changing Size redraws the panel. A setting that does not apply is never sent.

For an OpenRouter model, Choose photo keeps the photo on the phone instead of uploading it to Replicate. The notice reads "Keeping the photo on this phone. It goes to OpenRouter with each request.", the photo joins the Image feed, and the input holds it until it is removed. A photo uploaded to Replicate for another model cannot go to OpenRouter, and a photo kept for OpenRouter cannot go to a Replicate model; the panel says so before anything is sent. Refresh model inputs keeps the app's own list for OpenRouter and SpicyAPI models.

Pictures in a list input are labeled @img1, @img2, and so on, in the order they are sent. A description can name them that way, and the model receives "image 1", "image 2".

Each input of the model gets its own card, in the provider's order. Every card has one control, which shows what will be sent. An input starts at the model's default and is sent only after the user changes it. People who find ComfyUI and similar tools hard to read are the reason for this layout.

Box	What it holds	Why
Title	A plain name, such as "Safety checker", "Shape", or "Number of pictures", marked "(required)" when needed	The provider's own names, such as disable_safety_checker, read like code
Provider name	The input's exact name, in small gray type under the title	Matches the provider's documentation and the summary line on cards
Value	A list for fixed choices, a switch for yes or no, or a text field for numbers and text	One control per input. A list starts with "Model default", and an empty field means the model's default. Numbers use the number keyboard, and a comma typed as a decimal point is accepted.
Safety checker switch	"Safety checker on", on or off	It shows the checker itself, so on means on. The app sends the provider's reversed setting (disable_safety_checker) for you.
Use default	A link on a card the user has changed	Puts that input back to the model's default, so it is no longer sent
Helper line	The allowed range and the provider's description	Tells what the input does
Choose photo	A button on image and frame inputs, such as image, first_frame, and last_frame	Opens the phone's photo picker, uploads the photo to the user's Replicate account, and fills the input. Without it, the user would need a public web address for the photo.

Below the cards, a note lists any inputs the panel cannot edit, such as nested settings. A required input of that kind blocks the model, because the request would fail. Video models without a frame rate input show a note that points to the Frame rate filter.

Tools card

Box	What it holds	Why
Refresh model inputs	Reads the model's current inputs from Replicate (Image and Video)	Providers change their models. Refreshing keeps the input cards true to the live model. It is free and runs no generation.
Suggest setup with text LLM	Sends the prompt and the matching models' inputs to the chat model, then shows its suggestion for review	Helps choose a model and values for a prompt. The suggestion fills the panel only after the user taps Use in panel, and nothing generates until the user taps send. It costs one small chat request.
Reset this model's overrides	Puts every input of this model back to its starting value	Undoes experiments without hunting through every card

Bottom bar

Box	What it holds	Why
API keys	Leaves the panel and opens Settings for this mode	Fixes a missing key in two taps. Tap Apply first to keep panel changes, because this button does not save them. Back from Settings returns to the main screen.
Close	Leaves without saving	Same as the X, placed near the thumb
Apply	Saves the model and inputs, then returns to the main screen	The single commit point. It refuses a model that fails the active filters.