check_balance
ChatGPTShow how much VOSU balance is available and the current subscription state. Costs nothing. Use this before planning or generating, since ad planning costs 0.68 credits per scene. QUOTE THE RETURNED TEXT VERBATIM rather than restating creditsAvailable.
generate_3d
ChatGPTTurn an image into a textured 3D model (.glb). SPENDS FROM THE USER'S BALANCE (hundreds of credits per model — say so before calling). Works best on a single object with a clean background. Poll get_generation_status with the returned jobId (takes 2–6 minutes); the completed result carries glbUrl, a .glb model. Available models: - meshy-6 (Meshy 6.0) — clean, game-ready meshes with PBR textures (default) - hunyuan-3.1 (Hunyuan 3.1 Pro) — high-detail organic shapes and characters - tripo-3.1 (Tripo 3.1) — fast, low-cost drafts - meshy-7-multi (Meshy 7.0 Multi-View) — 2–4 photos of the same object from different angles (pass image_urls)
generate_image
ChatGPTGenerate an image from a text prompt (optionally editing a reference image). SPENDS FROM THE USER'S BALANCE. Available image models: - nano-banana-2 (Nano Banana 2) — highest quality, up to 4K, strong text rendering - nano-banana-lite (Nano Banana 2 Lite) — fast, cheap, good general quality - gpt-image-2 (GPT Image 2.0) — storyboards and conceptual scenes, faithful prompt following - z-image-turbo (Z Image Turbo) — fast, experimental, low-cost drafts - flux-2 (Flux 2) — precise prompt adherence and fine detail - seedream-v5-pro (Seedream v5 Pro) — up to 6K resolution and strong instruction editing - recraft-v3 (Recraft v3) — logos, icons, vectors, and mockups - grok-image (Grok Image) — expressive, high-contrast, stylized images - luma-photon (Luma Photon) — photographic realism - nano-banana-pro (Nano Banana Pro) — premium Gemini quality, up to 4K, precise edits - nano-banana (Nano Banana) — fast Gemini generation and quick edits - gpt-image-2.5-flare (GPT Image 2.5 Flare) — newest GPT Image, quality up to max, transparent bg - gpt-image-2.5-sunburst (GPT Image 2.5 Sunburst) — GPT Image 2.5 with a warmer, more stylized look - gpt-image-1.5 (GPT Image 1.5) — reliable prompt following at a lower cost - gpt-image-1 (GPT Image 1) — original GPT Image, general purpose - gpt-image-1-mini (GPT Image 1 Mini) — cheapest GPT Image, quick drafts - flux-2-pro (Flux 2 Pro) — production-grade photoreal Flux, strong detail - flux-2-max (Flux 2 Max) — highest-fidelity Flux, complex scenes - flux-2-flash (Flux 2 Flash) — very fast, low-cost Flux drafts - flux-2-turbo (Flux 2 Turbo) — fast Flux with better detail than Flash - flux-pro-kontext (Flux Pro Kontext) — in-context edits that keep subject and style - flux-pro-kontext-max (Flux Pro Kontext Max) — top-quality Kontext edits and typography - seedream-v5-lite (Seedream v5 Lite) — affordable Seedream 5, up to 3K - seedream-v4.5 (Seedream v4.5) — sharp detail and posters, up to 4K - seedream-v4 (Seedream v4) — low-cost Seedream, up to 4K - qwen-image-2-pro (Qwen Image 2 Pro) — text-heavy layouts, bilingual typography - qwen-image-2 (Qwen Image 2) — text in images at a lower cost - wan-2.7-pro (Wan 2.7 Pro) — cinematic, natural-looking images - wan-2.7 (Wan 2.7) — cinematic look at a lower cost - luma-photon-flash (Luma Photon Flash) — cheapest photoreal drafts - flux-dev (Flux Dev) — open-weight Flux, good all-round quality - flux-srpo-dev (Flux SRPO Dev) — Flux Dev tuned for more realistic skin and light - flux-lora (Flux Lora) — Flux Dev base; custom LoRAs are not supported here - flux-2-lora (Flux 2 Lora) — Flux 2 base; custom LoRAs are not supported here - flux-kontext-lora (Flux Kontext Lora) — Kontext edits; custom LoRAs are not supported here - flux-pro-v1.1 (Flux Pro v1.1) — fast, reliable Flux Pro quality - flux-pro-v1.1-ultra (Flux Pro v1.1 Ultra) — Flux Pro at up to 4MP, photoreal - minimax-image-1 (Minimax Image 1) — consistent characters from a subject photo; prompt max 1500 - ideogram-v3 (Ideogram V3) — readable text and logos; edits need image_url + mask_url - kling-image-o3 (Kling Image O3) — edit an existing image up to 4K (needs image_url) - kling-image-v3 (Kling Image 3.0) — edit an existing image up to 2K (needs image_url) - z-image-turbo-lora (Z Image Turbo Lora) — fast drafts with a style-tuned base - auto — auto-picks the best model for the prompt Pick the model whose 'best for' fits the request; if the user does not name one (or says 'auto'/'you pick'), choose the best fit and say which, or omit model for the default (gpt-image-2). If the user asks what image models are available, present the list above as bullets. For a model's settings (quality, resolution, background…) and what each costs, call model_options before setting options. A live preview appears and fills in with the finished image on its own — no need to poll.
generate_video_model
ChatGPTGenerate a video from a text prompt (optionally from a reference image). SPENDS FROM THE USER'S BALANCE at a per-second rate, so duration is required — the length, and therefore the price, must be something the user chose rather than a default they never saw. Available video models: - seedance-2.0 (Seedance 2.0) — strong character/identity consistency across the clip · durations: 4–15s · accepts video_url - seedance-2.5 (Seedance 2.5) — long continuous shots (up to 30s) driven by a reference video plus character/location images — the previz → cinematic workflow · durations: 4–30s · accepts video_url - kling-3.0 (Kling Video 3.0) — cinematic, high-quality motion and polished output · durations: 5, 10, 15s - veo-3.1 (Veo 3.1) — highest-quality video with native audio · durations: 4, 6, 8s - veo-3.1-lite (Veo 3.1 Lite) — fast and cheap iteration while drafting · durations: 4, 6, 8s - omni-flash (Omni Flash) — fast, low-cost general video generation · durations: 4–10s · accepts video_url - wan-3.0 (Wan 3.0) — following a reference video's camera move and layout (previz / Blender playblast → finished shot) · durations: 2–30s · accepts video_url - grok-video (Grok Video) — expressive, stylized motion · durations: 1–10s - seedance-1.5-pro (Seedance 1.5 Pro) — previous-gen Seedance; image mode needs image_url + end_image_url · durations: 4–12s - kling-o3 (Kling O3) — Kling's newest engine; start/end frames, up to 4k · durations: 5, 10, 15s - kling-o1-pro (Kling O1 Pro) — animate BETWEEN two frames — needs image_url AND end_image_url · durations: 3–10s - kling-o1-standard (Kling O1 Standard) — cheaper first→last frame animation — needs image_url AND end_image_url · durations: 3–10s - sora-2 (Sora 2) — native audio, strong physics, up to 20s · durations: 4, 8, 12, 16, 20s - sora-2-pro (Sora 2 Pro) — highest-quality Sora, 1080p · durations: 4, 8, 12, 16, 20s - pixverse-v6 (PixVerse v6) — effects-driven motion; start→end transitions · durations: 1–15s - vidu-q2 (Vidu Q2) — short clips; image mode needs image_url + end_image_url · durations: 2–8s - wan-2.6 (Wan 2.6) — music-video clips — REQUIRES a background-music audio_urls entry · durations: 5, 10, 15s - wan-2.5-preview (Wan 2.5 Preview) — legacy music-driven Wan — REQUIRES an audio_urls entry · durations: 5, 10s - minimax-h3-max (MiniMax H3 Max) — reference images + video pool, up to 1080p · durations: 5–15s · accepts video_url - minimax-hailuo-03 (MiniMax Hailuo 03) — 2K output, reference pool · durations: 5–15s · accepts video_url - flux-3 (FLUX 3 Video) — long clips (to 20s) with audio; start→end frames via routing · durations: 5–20s - also available (details & prices via model_options): seedance-2.0-fast (4–15s), seedance-2.0-mini (4–15s), seedance-1.0-pro (2–12s), seedance-1.0-pro-fast (2–12s), seedance-1.0-lite (2–12s), veo-3.1-fast (4|6|8s), kling-2.6-pro (5|10s), kling-2.5-turbo-pro (5|10s), kling-2.1-master (5|10s), kling-2.0-master (5|10s), kling-1.6-pro (5|10s), kling-1.6-standard (5|10s), kling-1.5-pro (5|10s), pika-2.2 (5|10s), pixverse-v5.5 (5|8|10s), pixverse-v5 (5|8s), pixverse-v4 (5|8s), pixverse-v4-fast, vidu-q3 (5s), ltx-2-distilled, ltx-2.3-pro (6|8|10s), wan-3.0-fast (2–30s), minimax-hailuo-2.3-pro, minimax-hailuo-2.3-standard (6|10s), minimax-hailuo-02-pro, minimax-hailuo-02-standard (6|10s), minimax-video-01, minimax-video-01-live, omni-flash-1.1 (3–10s) - auto — auto-picks the best model for the prompt Pick the model whose 'best for' fits the request; if the user does not name one (or says 'auto'/'you pick'), choose the best fit and say which, or omit model for the default (seedance-2.5). If the user asks what video models are available, present the list above as bullets. For a model's settings and what each costs, call model_options (kind: video). A live player appears and fills in on its own — no need to poll.
get_generation_status
ChatGPTCheck whether a video job has finished and get the video URL when it has. Costs nothing. Wait about 30 seconds between checks — rendering takes several minutes — and stop polling once the status is completed or failed. Tell the user it is still working rather than calling this repeatedly in one turn.
model_options
ChatGPTFree. Lists what a generation model accepts — its options keys and allowed values, aspect ratios, reference-image limits, prompt limit — and what each setting costs. Call it before generate_image or generate_video_model when the user wants a specific quality, resolution, transparent background or similar setting, or asks what something will cost. Omit model for a price overview of every model.
show_images
ChatGPTOpen an interactive gallery of the images this user has generated. Costs nothing. Use it whenever they ask to see, review, re-download, or reuse a past image ('show me my images', 'what have I generated'). Newest first. For images and videos together use show_media.
show_media
ChatGPTOpen an interactive gallery of ALL the user's generations — images and videos together, newest first. Costs nothing. Use it when they ask to see everything they've made ('show me all my generations', 'my whole library') without specifying images or videos. For images only use show_images, for videos only use show_videos. The user can play videos, open images fullscreen, and download from the gallery itself.
show_videos
ChatGPTOpen an interactive gallery of the standalone videos this user has generated with a video model. Costs nothing. Newest first. For images and videos together use show_media.
upload_media
ChatGPTOpen an upload panel so the user can add a local image or video from their device. Costs nothing. Call this whenever a file needs to become a hosted URL and you do not already have one — ESPECIALLY when the user ATTACHED or dragged an image into the chat: you can see that image but you CANNOT pass it to any generate tool (MCP tool arguments are plain JSON, with no way to carry a chat attachment), so this panel is the only way to turn it into a usable URL. Also use it for 'animate this picture', 'use my product shot', 'edit this photo'. Pass type: 'video' only if a video file is what is needed; the default is images. The panel handles the upload itself and sends you the hosted URL(s) when the user confirms. THEN: if the user already said what they want (a video, an edit, a product ad), do that with the URL — image_url / productImageUrls / images. If they only gave you the file with no instruction, ask what they'd like to make from it before spending.
