MiniMax H3 Max is MiniMax's officially launched high-speed H3 video model. EvoLink supports text-to-video, first-frame image-to-video with an optional last frame, and image/video/audio reference workflows at 480p or 768p for 5–15 seconds.
Test MiniMax H3 Max output, estimate credits, and pick the model ID that matches your input type before wiring the API into your product.
Create a Sample
Pick a route, enter your prompt, and set the parameters.Results are kept for 24 hours — download anything you want to keep.
All three routes charge for output seconds. Reference-to-video also charges for reference video seconds and images beyond the first two. Reference audio has no separate fee.
One 5s 480P sample
About 40 tests with $10 in credits.
The first test uses the cheapest visible route at 480p for 5s, with any required image input. Omitting quality defaults to 768p. Reference video and extra images add to the output charge.
Good for a first validation.
Good for prompt and shot iteration.
Good before production integration.
Estimates use the current per-second rate for your user group. Duration is rounded up to whole seconds, exactly as the task reports it.
| Mode | 480P | 768P | Billing |
|---|---|---|---|
| Text-to-VideoStandard route | $0.048/s-5% 3.23 cr/s$0.050 official price | $0.076/s-5% 5.168 cr/s$0.080 official price | Output video seconds |
| Image-to-VideoStandard route | $0.048/s-5% 3.23 cr/s$0.050 official price | $0.076/s-5% 5.168 cr/s$0.080 official price | Output video seconds |
| Reference-to-VideoStandard route | $0.048/s-5% 3.23 cr/s$0.050 official price | $0.076/s-5% 5.168 cr/s$0.080 official price | Output video seconds |
All three routes charge for output seconds. Reference-to-video also charges for reference video seconds and images beyond the first two. Reference audio has no separate fee.
Minimum charge per task equals the cheapest valid request (480p x 5s). Failed, expired, and rejected tasks are refunded in full.
| Item | Use Case | Price | Credits |
|---|---|---|---|
| Reference video480P | Applies when a Reference-to-Video request includes reference videos. Billed on top of the output cost. | $0.053/s-5% $0.056 official price | 3.5724 cr/s |
| Reference video768P | Applies when a Reference-to-Video request includes reference videos. Billed on top of the output cost. | $0.136/s-5% $0.143 official price | 9.2378 cr/s |
| Reference imagesFrom image #3 | Applies when a Reference-to-Video request sends more than 2 reference images. | $0.071 / image-5% $0.074 official price | 4.7804 cr |
The rate follows the output quality. Total reference video duration is rounded up to whole seconds before billing.
The first 2 reference images are free. Each additional image is billed at the same rate for both output qualities.
MiniMax H3 Max is MiniMax's officially launched high-speed H3 video model. EvoLink supports text-to-video, first-frame image-to-video with an optional last frame, and image/video/audio reference workflows at 480p or 768p for 5–15 seconds.
Each generation mode is its own model ID behind the same API key and credit balance - pick the route that matches your input.
minimax-h3-max-text-to-videoGenerate 768p or 480p video from a text prompt.
minimax-h3-max-image-to-videoAnimate a required first frame with an optional last frame.
minimax-h3-max-reference-to-videoUse image, video and optional audio references. At least one image or video is required; audio cannot be the only reference. Do not combine references with first/last-frame parameters.
Choose a route and output tier based on the input and delivery requirements of your workflow.
Video generation
MiniMax H3 Max is MiniMax's officially launched high-speed H3 video model. EvoLink supports text-to-video, first-frame image-to-video with an optional last frame, and image/video/audio reference workflows at 480p or 768p for 5–15 seconds.
Text, first frame with optional last frame, or reference media
Use image, video and optional audio references. At least one image or video is required; audio cannot be the only reference. Do not combine references with first/last-frame parameters.
minimax-h3-max-text-to-video / minimax-h3-max-image-to-video / minimax-h3-max-reference-to-video
MiniMax-H3-Max is the upstream model. EvoLink exposes three input modes through separate model IDs on one video endpoint and credit balance.
768p / 480p, 5-15s, MP4
768p is the default when resolution is omitted. 2K is not available on H3 Max, and 4-second clips are not supported.
Output seconds + reference input fees
All three routes charge for output seconds. Reference-to-video also charges for reference video seconds and images beyond the first two. Reference audio has no separate fee.
Async submit, poll or callback, no cancel
Tasks report pending / processing / completed / failed. Cancellation is not available, so failures refund instead.
These fields cover the main integration decisions. Open the API Reference for route-specific request and response examples.
modelstringChoose the model ID that matches text, first-frame or reference input.
Default: route-specific
promptstringDescribe the scene, motion, and camera behavior in up to 7000 characters.
Default: required
durationintegerSet 5 through 15 output seconds (4-second clips are not supported); billing follows the output duration.
Default: 5
qualityenumSelect 480p or 768p. 480p is billed at 5/8 of the 768p per-second rate; there is no 2K tier.
Default: 768p
aspect_ratioenumT2V defaults to 16:9 and supports six fixed ratios. I2V follows the first frame. R2V defaults to adaptive and also accepts the six fixed ratios.
Default: T2V: 16:9 · R2V: adaptive
image_startstringImage-to-Video: required HTTP(S) URL of the first frame.
Default: required (I2V)
image_endstringImage-to-Video: optional last-frame URL; a first frame is required.
Default: optional
image_urlsstring[]Reference-to-Video only: up to 9 images. The first two are free.
Default: []
video_urlsstring[]Reference-to-Video only: up to 3 videos, at least 2s each and 15s combined. Input video seconds incur a separate fee.
Default: []
audio_urlsstring[]Reference-to-Video only: up to 3 audio files, at least 2s each and 15s combined. Audio alone is not accepted and has no separate fee.
Default: []
callback_urlstringPublic HTTPS URL that receives the final task state; keep the task ID as a polling fallback.
Default: optional
Use the EvoLink API for product integration and batch jobs, or call it from Codex, Claude, or Gemini for fast creative and development workflows. Both paths share the same EvoLink API key, balance, model routes, and task history.
Best for: product backends, batch jobs, automated workflows
Call EvoLink’s unified video API from your server and control the model ID, parameters, task queue, callbacks, and result storage.
Best for: creative and development tasks in Codex, Claude, and Gemini
Give the Agent your output goal, assets, and acceptance criteria. It can choose the route, assemble the request, track the task, and return the result without requiring you to hand-code every step.
Use text prompts, a controlled first and last frame, or image, video and audio references across creative, commerce, and production workflows.
This example shows the shortest runnable flow: create a task, save the returned task ID, then poll or wait for an HTTPS callback. Open the API tab for the complete parameter and response reference.
curl -X POST https://api.evolink.ai/v1/videos/generations \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-h3-max-text-to-video",
"prompt": "A female captain before a vast observation window as the fleet jumps away, cinematic slow dolly-in",
"duration": 5,
"quality": "768p",
"aspect_ratio": "16:9",
"callback_url": "https://your-domain.com/webhooks/video-task-completed"
}'
# Save the returned task ID, then query:
curl https://api.evolink.ai/v1/tasks/{task_id} \
-H "Authorization: Bearer $EVOLINK_API_KEY"Check every value against the parameter cards above: ranges, allowed options, asset counts, and file types must match the selected route.
Check the Authorization bearer token and confirm the available balance and task billing in the console.
Review likeness rights, brand assets, readable text, sensitive content, and input-asset compliance.
A MiniMax H3 Max task that reaches a final failed state is not charged. Keep the task ID, inspect the final state, and identify whether the cause is validation, review, or execution before retrying.
callback_url must be a public HTTPS URL, not localhost or a private network address. Always keep the task ID as a polling fallback.
Choose H3 Max for fast 480p / 768p iteration; switch routes when you need higher resolution or broader reference inputs.
Use real workloads and acceptance criteria — not selected showcase clips — to measure usable output, consistency, total cost, and operating reliability.
Measure prompt adherence, temporal consistency, visible artifacts, and accepted outputs across the full attempt set.
Compare character, face, clothing, logo, product shape, composition, and style retention from the same source assets.
Include queue limits, failed-task refunds, moderation, retries, reviewer correction, delivery format, and downstream editing — not only generation speed.
Calculate cost per final accepted clip, and confirm callbacks, retention rules, monitoring, rollback, and multi-model fallback fit your product.
| Feature | MiniMax H3 Max | MiniMax H3 | Other video routes |
|---|---|---|---|
| Primary role | Fast Hailuo route for volume | Full-featured Hailuo route | Cross-vendor comparison and fallback |
| Generation modes | T2V + I2V + R2V | T2V + I2V + R2V | Model-dependent |
| Reference input | Image + video + audio (12 items max) | Image + video + audio | Model-dependent |
| Resolution | 768p / 480p | 2K / 768p | 480p to 4K |
| Duration | 5-15s | 4-15s | 3-15s |
| Billing | Output seconds + reference input fees | Per output second (+ reference video seconds) | Per second or per video |
Pick the model ID that matches your input: minimax-h3-max-text-to-video for a prompt alone, minimax-h3-max-image-to-video for a required first frame with an optional last frame, and minimax-h3-max-reference-to-video for image, video and audio references.
Create the task once, then poll its status or receive a callback until the MP4 is ready. Result URLs remain available for 24 hours, so copy completed videos to your own storage as part of the delivery flow.
Use 480p for drafts and 768p for higher-definition output. Include reference input fees in the budget; use H3 when 2K is required. Failed tasks are refunded.
See credits and per-second cost before each generation.
Switch between Hailuo, Seedance, Kling, and other video routes without rebuilding auth or billing.
Track credits, spending, and model usage for every route in one console.
See whether a task is pending, processing, completed, or failed, with refunds on failure.
Keep MiniMax H3 Max as one tested route and fall back to H3 or another family when a workload needs it.
Compare Hailuo generations under one EvoLink account and API key, then choose by resolution, duration, reference workflow, and cost.
MiniMax H3 Max Turbo
Lower-cost text-to-video and first-frame image-to-video at 480p / 768p, with 5–15s output.
View
MiniMax H3
The full-featured Hailuo route: 2K / 768p output, 4-15s, and text, frame, or reference driven generation.
View
Hailuo 2.3
The documented Hailuo generation for text-to-video and image-to-video through Standard and Fast routes.
View
Seedance 2.0
A multimodal route for text, image, video, and audio reference workflows from 480p to 4K.
View
Kling O3
A production video route with multimodal controls and a different capability and cost profile.
View
Wan 2.7
A video fallback for text-to-video and image-to-video workflows with different duration and pricing options.
View
Veo 3.1
Google DeepMind's Veo 3.1: 8-second clips with native audio, available in Fast and Pro variants.
ViewUse the family page and the dated comparisons for the decision that brought you here.
cURL, Python, and TypeScript examples for the Hailuo routes, async submission, polling, callbacks, and error handling.
Confirmed evidence, reported timing, and the verification checklist EvoLink used before opening the Hailuo routes.
Decide when the next generation has enough evidence to justify an upgrade inside the same family.
Compare the Hailuo routes with an available multimodal audio-video workflow.
Decide whether to ship on Hailuo or move to the Seedance 2.5 API, live on EvoLink since August 7, 2026.
MiniMax H3 Max is MiniMax's officially launched high-speed H3 video model. EvoLink supports text-to-video, first-frame image-to-video with an optional last frame, and image/video/audio reference workflows at 480p or 768p for 5–15 seconds.
H3 Max offers 480p / 768p output for 5–15s. H3 offers 768p / 2K for 4–15s. Both support text, frame and reference workflows, with different reference input fees.
High-volume advertising creative, ecommerce product video, social clips, and fast storyboard checks. Start from a text prompt, drive a shot with a required first frame and an optional last frame, or guide subject and style with image, video and audio references — any workflow where iteration speed matters more than the 2K tier.
MiniMax-H3-Max is the upstream model. EvoLink exposes three input modes through separate model IDs on one video endpoint and credit balance.
minimax-h3-max-text-to-video / minimax-h3-max-image-to-video / minimax-h3-max-reference-to-video. Choose the model ID that matches text, first-frame or reference input.
All three routes charge for output seconds. Reference-to-video also charges for reference video seconds and images beyond the first two. Reference audio has no separate fee.
Use image, video and optional audio references. At least one image or video is required; audio cannot be the only reference. Do not combine references with first/last-frame parameters. Up to 9 images, 3 videos, 3 audio files; 12 total.
No on EvoLink's current H3 Max routes. They accept 480p and 768p, and reject 2K. Use MiniMax H3 when you need the 2K tier.
Five seconds. EvoLink's H3 Max contract accepts 5-15 seconds and rejects 4-second clips. The minimum charge follows the same floor: 480p at 5 seconds.
No. Output is 768p or 480p with durations of 5 to 15 seconds, and there is no cancel endpoint, so tasks report can_cancel false and failures are refunded in full.
Submit the task, then poll the task endpoint or register a callback. When the status turns completed the response carries the MP4 URL. Download it and store it yourself.
Results are kept for 24 hours. Save anything you want to keep within that window — the URL stops resolving afterwards.