Generation
Use these tools to generate media. They spend credits, except where noted. Each one returns a job_id at once, and get_job_status reports its progress.
Pictures, clips and audio
Section titled “Pictures, clips and audio”generate_scene_image
Section titled “generate_scene_image”Generate a scene’s still from a text prompt, for one scene or a batch of up to 20 in a single call. Not on ChatGPT app v2; use generate_image there.
- Surfaces: ChatGPT app, Generic MCP
- Parameters:
project_id(required),aspect_ratio,negative_prompt,prompt,scene_id,scenes
refine_scene_image
Section titled “refine_scene_image”Edit a scene’s still with an instruction, optionally guided by up to 14 reference pictures. Not on ChatGPT app v2: use refine_image there.
- Surfaces: ChatGPT app, Generic MCP
- Parameters:
edit_instruction(required),project_id(required),scene_id(required),reference_image_urls,reference_images,reference_resource_ids,resource_id
generate_image
Section titled “generate_image”Generate a picture from a text prompt: a scene’s still, a picture for a reference document, or a standalone picture. Make one at a time, or a batch of up to 20 that can mix all three. It takes up to 14 reference pictures.
On ChatGPT app v2 and in Manan, reference_document_ids names the reference documents a generation must match. A recurring character, object or place then stays consistent across scenes. Each document attaches as its sheets, or as one grid of its pictures when it has no sheet. Screel writes the references into the scene’s prompt as mention tokens, so they show in the editor and survive a regenerate.
On ChatGPT app v2 it also takes chat file attachments as reference_images, and reference_image_urls name only pictures Screel already stores. A batch item can name its own reference_image_urls, such as the previous scene’s still for continuity. A picture that lands in a document takes an image_name such as front or side.
- Surfaces: ChatGPT app v2, Generic MCP, Manan
- Parameters:
project_id(required),aspect_ratio,document_id,image_name,images,negative_prompt,prompt,reference_document_ids,reference_image_urls,reference_images,request_id,scene_id,scene_position,target_bit_id
refine_image
Section titled “refine_image”Edit an existing picture with an instruction (a scene’s still, a document’s picture or a standalone picture), one at a time or batched. It takes the same reference pictures as generate_image, and on ChatGPT app v2 and in Manan the same reference_document_ids. With a product document, a scene refine places the product into the scene and keeps the rest.
In Manan, a still captured from Unreal also takes mask_object_ids, the stencil ids of the objects to change. The rest of the still stays as it is. It can also edit one keyframe of a scene (keyframe_id). With all_keyframes it edits the still and every keyframe with one instruction. It edits the still first, so the keyframes can follow its look. scope narrows that to the out-of-date or the recaptured pictures. scene_ids or scene_positions cover several scenes in one call.
In Manan, from_raw starts from the raw capture of an Unreal still or keyframe. It also starts from the raw frame of a still or keyframe that import_footage made. Screel makes a clay copy, describes the frame and repaints it in the look the instruction names. On a frame with no Unreal capture, Screel also offers the project’s reference documents that hold a sheet. It attaches only the sheets of the subjects in the frame. describe and clay turn those two steps off, and saved_prompt runs the scene’s saved refine prompt again.
On every MCP surface, a scene refine edits the picture as it is.
- Surfaces: ChatGPT app v2, Generic MCP, Manan
- Parameters:
project_id(required),document_id,edit_instruction,image_id,image_url,images,reference_document_ids,reference_image_urls,reference_images,request_id,resource_id,scene_id,scene_position,target_bit_id,target_image_id
animate_scene
Section titled “animate_scene”Turn a scene’s still into a clip of 3 to 15 seconds from a motion prompt, for one scene or a batch. resolution picks the tier: 360p draft by default on the ChatGPT app, or 720p, 1080p and 2160p.
On ChatGPT app v2, end_frame makes the clip end on a picture, so consecutive scenes flow into each other. It takes next, a scene id or a Screel picture url.
On ChatGPT app v2 and in Manan, reference_document_ids sends the documents of the products and props that a clip shows. Only a document whose sheets are object sheets goes with a clip; people and places come from the still.
In Manan and on ChatGPT app v2, lip_sync makes the person in the still sing the part of the music track under that scene. The clip runs the scene’s length. The project must play its music track as the music master.
- Surfaces: ChatGPT app, ChatGPT app v2, Generic MCP, Manan
- Parameters:
project_id(required),aspect_ratio,draft,duration,end_frame,generate_audio,lip_sync,motion_prompt,reference_document_ids,request_id,resolution,scene_id,scenes
generate_voiceover
Section titled “generate_voiceover”Generate a scene’s narration, for one scene or a batch, with a voice, a style and a speed. A line holds up to 200 spoken characters.
- Surfaces: ChatGPT app, ChatGPT app v2, Generic MCP, Manan
- Parameters:
project_id(required),request_id,scene_id,scenes,speed,style_key,text,voice
generate_music
Section titled “generate_music”Generate a background music track for the project from a prompt. The track length depends on the account: get_account reports it.
- Surfaces: ChatGPT app, ChatGPT app v2, Generic MCP, Manan
- Parameters:
project_id(required),prompt(required),duration_seconds,request_id
upscale_render
Section titled “upscale_render”Upscale a finished render to 1080p, 1440p or 2160p. Not available on every account: get_account reports whether it is.
- Surfaces: ChatGPT app, ChatGPT app v2, Generic MCP, Manan
- Parameters:
project_id(required),render_id(required),resolution(required),request_id
list_voices
Section titled “list_voices”The available narration voices and style presets, each with a one-sentence description, plus the defaults. It spends no credits.
- Surfaces: ChatGPT app, ChatGPT app v2, Generic MCP, Manan
- Parameters: none
Reference sheets
Section titled “Reference sheets”A sheet document’s views come from one picture: the subject repeated at the same scale on one plain backdrop.
generate_sheet
Section titled “generate_sheet”Every view of a sheet document from one high-resolution picture, made from a design paragraph and optional reference photos. Screel cuts the picture into one bank picture per view, and these views replace the views the sheet held. The earlier views move to a gallery below the sheet. An object has left side, right side, top and front three-quarter views. A character has front, three-quarter, side and back views, and a location has wide, entrance, interior and detail views. You can also ask for your own views. In a document that holds one sheet per subject, block_id names the sheet block to fill; left empty, the first one.
On ChatGPT app v2 it takes your photos as chat file attachments (reference_images) and pictures Screel already stores (reference_image_urls). A product packshot sheet from your photos is one use.
In Manan, look_document_id names a look document, and the views follow its look instead of photoreal. There, frame_url names a frame of your film that shows the subject, and the views draw that exact subject. It spends credits at the picture rate. With frame_url, it costs the same as a refine from a raw frame.
- Surfaces: ChatGPT app v2, Manan
- Parameters:
design(required),document_id(required),project_id(required),block_id,canonical,reference_image_urls,reference_images,reference_notes,request_id,views
