account_credit
ChatGPTQuery the available credit balance for the authenticated account.
account_models
ChatGPTQuery available AI models and their parameters. Types: 1501=Image-to-Video, 1502=Text-to-Video.
audio_tts
ChatGPTConvert text to speech with a voice from voice_list. Async: poll voice_result with the returned _id until data.result.status is 3; the audio URL is data.result.preview. Consumes credits per character; Free plan max 1000 characters.
avatar_detail
ChatGPTGet detailed information about a specific avatar by its ID.
avatar_list
ChatGPTGet a paginated list of available talking avatars. Can filter by type and source.
facedetect_analyze
ChatGPTAnalyze multiple image frames to identify and deduplicate persons across frames. Useful for building face_mapping for faceswap_plus multi-face workflows.
facedetect_detect
ChatGPTDetect faces in an image or video URL. Returns 6-point landmarks, bounding boxes, and optionally cropped face image URLs.
faceswap_delete
ChatGPTDelete face swap result records by comma-separated IDs.
faceswap_image
ChatGPTSwap a single face in an image using Face Swap Pro V4 — highest quality. Provide one source face and one target image.
faceswap_plus
ChatGPTRecommended face swap — supports single or multi-face, image and video targets. Uses akool_faceswap_classic_plus model.
faceswap_result
ChatGPTQuery face swap results by comma-separated IDs.
file_upload
ChatGPTGet a one-time upload for a local image, video or audio file, for agents that can run shell commands (Claude Code, Codex). Run the returned curl after replacing <LOCAL_FILE_PATH>; it prints 204 on success. Then pass file_url to other tools. On Windows use curl.exe. claude.ai web cannot upload local files: use a public URL instead. Accepted: .jpg .jpeg .png .mp4 .mov .avi .mp3 (convert webp/wav/m4a first). The upload link is valid for 7 days.
image_create
ChatGPTGenerate images from a text prompt (text-to-image), or from a source image URL plus a prompt (image-to-image). Async: data is one image record, or an array when count > 1; poll image_info with each record _id. Credits per image: 3 at 1080p, 4 at 4k.
image_info
ChatGPTGet an image generated by image_create, by its _id. image_status: 1=queued, 2=processing, 3=done, 4=failed (see error_reasons). At status 3 the image URL is in the image field. Poll every 5-10 s.
show_result
ChatGPTShow finished Akool results (video, image or audio) to the user in the chat, with a player and a Download button. Call once, after polling shows every result is done, with all final result URLs in viewing order. Skip intermediate files such as TTS audio that only feeds another tool. Only Akool result URLs (the CDN URLs returned by Akool tools) are accepted. Pass models with the model's display name (the account_models label, e.g. "Kling 2.6") for each result whose tool had a model choice (text-to-video, image-to-video, motion swap), '' otherwise; the card shows it next to the aspect ratio.
video_analyze
ChatGPTCall video_edit_guide first. Break a video on the Akool CDN into shots, flashes, pacing, loudness and beats, with storyboard contact sheets. An audio file (mp3 etc.) gives only loudness and beats. Returns the analysis JSON and up to 4 storyboard images, or a job_id to poll with video_edit_result. Consumes credits; failures are refunded.
video_edit_guide
ChatGPTRead before any other video_* edit tool. Akool's video editor, for when the user asks to trim or shorten a video to an exact length, cut, join clips, add music, voice-over, captions or text, resize, or remake a reference video with their own material. Returns the workflow, the JSON edit spec format, limits, credits and copyright rules.
video_edit_result
ChatGPTCall video_edit_guide first. Status of a video_analyze or video_render job: running, failed (credits refunded) or done. Done: the analysis or the render report (video_url, metrics, warnings) plus up to 4 contact sheets (storyboards or QC pages) as images; pick others with pages. Look at the QC pages before telling the user it is finished; show the final video with show_result.
video_frames
ChatGPTCall video_edit_guide first. Contact sheets of the frames in [start, end) of a video on the Akool CDN, each tile labelled with frame number and time. Use it to check fast cuts, flashes, transitions and on-screen text (crop to zoom in). Free, rate limited.
video_image2video_create
ChatGPTGenerate a video from a source image with a text prompt describing the motion. Pick model_name, resolution and video_length from account_models with types [1501]. Async: poll video_result with each data.successList[]._id. Paid plans only; consumes credits.
video_image2video_delete
ChatGPTDelete video generation records by comma-separated IDs (max 100).
video_image2video_effects
ChatGPTGet available special effects that can be applied during image-to-video generation.
video_image2video_results
ChatGPTQuery video generation results by comma-separated IDs (max 100).
video_image2video_update_audio
ChatGPTAdd or replace audio on a completed image-to-video result. Omit audio_url to let AI generate a soundtrack. Async: returns a new video _id to poll with video_image2video_results. Consumes credits.
video_motion_swap_create
ChatGPTThe person in each image performs the motion of the driving video: one video per image, as long as the driving video. Take each model's top-level model_name (not params.model_name) and an exact params.resolution.enum value (case-sensitive) from video_motion_swap_models; pass only options that appear in the params of that model. Cost per image ≈ ceil(video seconds / unit_duration) × unit_credit from unit_price_map (list price). Async: poll video_result with each data.list[]._id; data.errors lists images that failed to start, by index. Local files: upload with file_upload first.
video_motion_swap_models
ChatGPTList motion swap models (data.list). Call before video_motion_swap_create. For each model use its top-level model_name (not params.model_name); pass resolution as an exact value from params.resolution.enum (case-sensitive); pass only the options that appear in params (mode, character_orientation, keep_original_sound, prompt); params.image and params.video describe the input limits. List price per image = ceil(video seconds / unit_duration) × unit_credit, from unit_price_map[resolution], else unit_price_map.default. Ignore sale_unit_price. A model the account plan cannot use fails at create with "Your current account does not support this model".
video_render
ChatGPTCall video_edit_guide first. Render a JSON edit spec into an MP4 on Akool's servers (trim, join, layer video, images, text and captions, mix music and voice). The spec is checked first: if it has errors they come back as a list and nothing is charged; fix them all and call again. Otherwise returns a job_id; poll video_edit_result. Use draft for quick checks. Consumes credits; failures are refunded.
video_result
ChatGPTGet the status of any video task by its _id. video_status: 1=queued, 2=processing, 3=done, 4=failed (see error_reason). Poll every 15-30 s; the video URL appears at status 3 and expires after about 7 days.
video_talking_avatar_create
ChatGPTMake an avatar speak, full frame on a black background at its own aspect ratio (short side 1080). Pick the avatar from avatar_list and copy its avatar_id, url (as avatar_url) and from (as avatar_from). Speech: text + voice_id (from voice_list) or audio_url, exactly one. If text fails with "create talkingavatar error", make the audio with audio_tts, poll voice_result until data.result.status is 3 and pass data.result.preview as audio_url. Async: poll video_result with data._id. Credits scale with audio length; the text length limit depends on the plan.
video_talking_photo_create
ChatGPTAnimate a photo so it speaks an audio track. Needs an audio URL, there is no text input: to make it say text, first call audio_tts, poll voice_result until data.result.status is 3, and pass data.result.preview as audio_url. Async: poll video_result with data._id. Credits per started 5 s of audio: 720p 10, 1080p 15, 2k 20, 4k 25.
video_text2video_create
ChatGPTGenerate a video from a text prompt. Pick model_name, resolution and video_length from account_models with types [1502] (value, resolutionList, durationList). Async: poll video_result with each data.successList[]._id. Paid plans only; credits depend on model x duration x resolution.
voice_list
ChatGPTList voice models for text-to-speech, paginated. Returns { result, count, page, size }; use a voice_id with audio_tts.
voice_result
ChatGPTGet a text-to-speech result by the _id from audio_tts. data.result.status: 1=queued, 2=processing, 3=done, 4=failed. At status 3 the audio URL is data.result.preview (pass it as audio_url to other tools). Poll every 5-10 s.