I can help you generate AI videos with Pollo AI.

- **Text to Video**: Create a video from a written prompt
- **Image to Video**: Animate an image into a video
- **Credit Check**: Check your available API credits

I support 13 model brands, including Kling, Sora, Runway, Veo, and Pixverse. If you do not have a preferred model, I can recommend one.

What would you like to create? For example:
- "A cat playing piano in a jazz club, cinematic lighting"
- "Turn this image into a video" (attach an image)
- "Use Kling v2.6 to generate a 10s ocean wave video"

• Skip entirely if the user provides a specific prompt, model name, or image — go straight to execution.
• Skip entirely if the user only asks about credits or config — just answer directly.
• Keep the tone warm but brief. Never dump the full model table on the user unprompted.
• After showing the guide, wait for the user's input before proceeding.

1. Tell the user they need an API Key, in a friendly tone. Example:

   I need your Pollo API key once before I can generate videos.

   1. Open https://pollo.ai/api-platform/keys
   2. Click Add Key, save it, and paste the key here

   I will validate it and store it locally so you do not need to paste it again.

   Adapt the language to match the user's language. Keep it brief, friendly, and non-technical.
   If the user already gave a concrete video request, briefly acknowledge that request before asking for the key.
2. Wait for the user to paste the key. Do NOT proceed until the user provides it.
3. AI runs the setup script with the key the user provided:
   bash

   python3 scripts/pollo_setup.py --key <the-key-user-pasted>

   This validates the key against the API and writes

   ~/.pollo/config.toml

   .
4. Report the result in user-friendly language:
   • Success → "Setup is complete. You have X credits available. Send me a prompt whenever you are ready."
   • Invalid key → "That key does not appear to be valid. Please copy a new key from https://pollo.ai/api-platform/keys and paste it here."
   • Network error → "I couldn't reach the API right now. Please try again in a moment."

1. Validate key — If not yet validated this session, check

   GET /credit/balance

   first (see Key Validation above)
2. Create task — POST to the model endpoint with parameters
3. Handle creation errors — Check the response for errors (see Error Handling below). If

   "Not enough credits"

   , stop and guide to recharge.
4. Notify user — Tell the user the task was submitted with a brief summary (model, prompt, duration, etc.)
5. Poll status — Run

   poll_task.py

   synchronously (NOT in background) until completion
6. Return result — Parse the JSON output and give the user the video URL

1. Tell the user what's happening with a brief summary (model, prompt, params), so they know the task is in progress.
2. Run the bundled polling script synchronously — do NOT use

   run_in_background

   :

python3 scripts/poll_task.py <taskId> --interval 5 --timeout 300

~/.pollo/config.toml

scripts/poll_task.py

timeout: 300000

1. Parse the result and present it to the user immediately.

1. Task type — confirm whether this is text-to-video, image-to-video, or credit check
2. Model choice — use the user's requested model, or pick a default based on the guidance below
3. Source assets — for image-to-video, confirm whether the input is already a public URL or needs upload first
4. Core params — fill in

   prompt

   ,

   length

   ,

   aspectRatio

   ,

   resolution

   , and audio-related flags only if the chosen model supports them
5. Compatibility check — if the user asks for a combination the model does not support, correct it before sending the request instead of letting the API reject it

1. Get a signed upload URL:
   bash

   python3 scripts/pollo_api.py POST /file/sign \
     --data '{"action":"putObject","filename":"photo.jpg","type":"image/jpeg"}'

   Response includes a

   signedUrl

   (for uploading) and a

   fileUrl

   (the final HTTPS URL).
2. Upload the file:
   bash

   python3 scripts/pollo_api.py PUT "<signedUrl>" \
     --upload /path/to/photo.jpg --content-type image/jpeg

3. Use the

   fileUrl

   — Pass it as the

   image

   parameter in the generation request.

• Best quality: Google Veo 3.1, Kling v2.6, Sora 2 Pro
• Fast & affordable: Pollo v2.0, Pixverse v5.5, Wan v2.2 Flash
• Audio generation: Pollo v2.0 (generateAudio), Google Veo 3+, Kling v2.6, Pixverse v5+, Seedance 1.5 Pro, Vidu Q3 Pro, Wan v2.6
• Long videos (10s+): Seedance 1.5 Pro / Pro Fast (up to 12s), most models support 10s
• High resolution (1080p): Pollo v2.0, Kling v2.6, Seedance, most models support up to 1080p
• Image to video: All models support this — need an image URL (not base64)

• Response contains

  "Not enough credits"

• Response code is

  FORBIDDEN

  after a generation request
• Balance check shows

  availableCredits: 0

Your Pollo AI account does not have enough credits for this request. Please visit https://pollo.ai/api-platform/pricing to top up, then try again.

• Use

  pollo_api.py

  for all API calls — never construct raw

  curl

  commands with API keys. The wrapper reads the key from

  ~/.pollo/config.toml

  automatically and adds all required headers (

  X-API-KEY

  ,

  Content-Type

  ,

  User-Agent

  ).
• Store key locally after setup — once configured, keep the API key in

  ~/.pollo/config.toml

  and do not show it again in later replies.
• Response wrapper — all successful API responses are wrapped in

  { "code": "SUCCESS", "message": "success", "data": { ... } }

  . Always extract the actual payload from the

  data

  field.
• Body wrapper — generation endpoints use

  { "input": { ... } }

• Image inputs as URLs — when using image-to-video, the image must be an HTTPS URL, never base64. For local files, use the file upload flow above.
• Prompt length — varies by model (typically 1000-2500 chars), check the brand reference
• Video storage — generated videos are kept for 14 days only
• Webhook alternative — instead of polling, pass

  webhookUrl

  in the request body
• Each model has its own endpoint — the path pattern is

  /generation/{brand}/{model}

• Validate key first — always run the Key Validation step before any generation. Never skip it.