SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillskie-ai

kie-ai

General & other

View repository

Updated 10-16 07:22

About this skill

kie-ai — Grok Imagine via Kie.ai API

Generate image/video via Kie.ai REST API — proxy for Grok Imagine models

Path Shortcuts

OUT    = D:/ClaudeMediaGen/output-fal   # ใช้ร่วมกับ fal output
BASE   = https://api.kie.ai/api/v1/jobs

Environment Variables

Var Address How to read
KIE_API_KEY Windows User Env os.environ["KIE_API_KEY"] (set via setx or [Environment]::SetEnvironmentVariable)

Model Selection

Model ID Type Best For Output
grok-imagine/text-to-image text2img Aesthetic/cinematic images, good with Thai prompts 3 รูป/request
grok-imagine/image-to-image img2img Edit/adjust style from a reference image TBD
grok-imagine/image-to-video img2vid Create video from a still image TBD
grok-imagine/upscale upscale Increase video resolution TBD
veo3_fast video Video fast, supports i2v (REFERENCE_2_VIDEO) video + audio
veo3_quality video Video highest quality video + audio
veo3_lite video Video most economical video + audio

Note: text-to-image + veo3_fast tested and actually works / other models are in docs but not yet tested Veo3 uses a different endpoint — see section "Veo3 Video Generation" and "API Endpoints Summary" below

Aspect Ratios

Value Use when
2:3 Portrait (default for people images)
3:2 Landscape
1:1 Square
16:9 Wide cinematic
9:16 Story/Reels format

API Flow — 3 steps

1. Create Task

POST {BASE}/createTask
Authorization: Bearer <KIE_API_KEY>
Content-Type: application/json

Body:
{
  "model": "grok-imagine/text-to-image",
  "input": {
    "prompt": "...",
    "aspect_ratio": "2:3"
  }
}

Response: { "code": 200, "data": { "taskId": "..." } }
  • Check code == 200 first; if not, = error
  • Keep taskId for polling

2. Poll Task Status

GET {BASE}/recordInfo?taskId=<taskId>
Authorization: Bearer <KIE_API_KEY>

Task States:

State Meaning Action
waiting In queue Keep waiting
queuing In queue Keep waiting
generating Generating Keep waiting
success Done ดึง URL
fail Failed Read failMsg + failCode

Poll strategy: every 5 seconds, timeout 5 minutes (60 rounds)

When success:

  • data.resultJson is a JSON string → must parse it again
  • Inside there is resultUrls: [url1, url2, url3] — gets 3 images per 1 request

3. Download Images

Important: URL must include special headers, otherwise it will be blocked

Headers สำหรับ download:
  User-Agent: Mozilla/5.0
  Referer: https://kie.ai/

Naming convention:

kie_grok_{taskId_last8}_{index}.jpg

Output directory: D:/ClaudeMediaGen/output-fal/

Prompt Rules

Grok prompt syntax rules have moved to the shared model-card. See grok-image.md for:

  • Hybrid prompt strategy (English + Thai)
  • Forbidden words
  • Subject short / Scene long rule
  • Do not specify body size

Implementation Guide

Create a gen script in Python

  1. Read API key from os.environ["KIE_API_KEY"] (Windows User Env — set once system-wide)
  2. Create task → POST /createTask → keep taskId
  3. Poll → GET /recordInfo?taskId= → loop every 5s until getting success or fail
  4. Parse result → data.resultJson as a string → json.loads() → resultUrls
  5. Download → put in User-Agent + Referer headers → save as .jpg
  6. Create a paired .txt → save the prompt text alongside the image every time (for traceability)

Batch Generation

  • Multiple requests can be sent simultaneously (no rate limit encountered in testing)
  • Each request yields 3 images → 10 requests = 30 images
  • Use run_in_background: true for parallel gen in Claude Code

Error Handling

  • code != 200 from createTask → show the full error response
  • state == "fail" → read data.failMsg + data.failCode
  • Network error while polling → just retry, no need to re-create the task
  • Download fail → retry with headers

Cost

  • No cost per request data from Kie.ai yet — always use preview before batch

Rules

  • Preview before batch — gen 1 request first (yields 3 images), check quality before generating more
  • Do not Read generated images — wastes a lot of context, just report the file path
  • .txt always paired — save the prompt text alongside the image every time
  • Do not Read the output file of a background task — issue the gen command and be done

File Upload API (tested)

Use to upload files (e.g. images for image-to-video) before submitting gen

POST https://kieai.redpandaai.co/api/file-stream-upload
Authorization: Bearer <KIE_API_KEY>
Content-Type: multipart/form-data

Parameters:
  file: (binary)
  uploadPath: (string)
  fileName: (string)

Response:
{
  "success": true,
  "data": {
    "downloadUrl": "https://tempfile.redpandaai.co/..."
  }
}

Note:

  • Use curl multipart only (base64 upload returns 403)
  • Files are automatically deleted after 3 days
  • Must upload before using image-to-video (Veo3 REFERENCE_2_VIDEO)

Veo3 Video Generation (tested)

Veo3 uses a different endpoint from the Market API (Grok Imagine) — both generate and poll

Create Task

POST https://api.kie.ai/api/v1/veo/generate
Authorization: Bearer <KIE_API_KEY>
Content-Type: application/json

Body:
{
  "prompt": "...",
  "imageUrls": ["https://uploaded-url"],
  "model": "veo3_fast",
  "generationType": "REFERENCE_2_VIDEO",
  "aspect_ratio": "9:16"
}
Parameter ค่า Note
model veo3_fast, veo3_quality, veo3_lite fast supports i2v
generationType TEXT_2_VIDEO, REFERENCE_2_VIDEO REFERENCE_2_VIDEO works only with veo3_fast
aspect_ratio 16:9, 9:16, Auto —
imageUrls array of URLs Must upload first via file-stream-upload

Optional: enableTranslation, enableFallback, watermark, callBackUrl, seeds

Poll Status

GET https://api.kie.ai/api/v1/veo/record-info?taskId=<taskId>
Authorization: Bearer <KIE_API_KEY>
Field ค่า Meaning
successFlag 0 Generating
successFlag 1 Success
successFlag 2 Failed
successFlag 3 Failed
  • Video URL is in data.response.resultUrls[]
  • มี audio (hasAudioList: [true])
  • Takes approximately 60-120 seconds

Veo3 Flow: Upload → Generate → Poll

1. Upload image  → POST file-stream-upload → ได้ downloadUrl
2. Generate      → POST /veo/generate (ใส่ downloadUrl ใน imageUrls) → ได้ taskId
3. Poll          → GET /veo/record-info?taskId= → วน loop จน successFlag != 0
4. Download      → ดึง video จาก resultUrls[]

API Endpoints Summary

Feature Market API (Grok Imagine) Veo3 API
Create task POST /api/v1/jobs/createTask POST /api/v1/veo/generate
Poll status GET /api/v1/jobs/recordInfo GET /api/v1/veo/record-info
File upload POST kieai.redpandaai.co/api/file-stream-upload (same)
Base URL https://api.kie.ai https://api.kie.ai

Related Skills

  • /art-engine - upstream: model selection and prompt syntax translation
  • /fal-ai - alternative execution path for Grok (same model, different API - fal.ai direct vs Kie.ai proxy)
  • /art-director - upstream: creative brief crafting
  • /sira-image-prefer - taste DNA for prompt preferences