What Hailuo 03 is best at
Four tiers, one prompt style
Hailuo 03 tops out at 2K and carries references and trained models. H3 Max renders 480p, 768p, or 1080p with stronger prompt adherence and higher throughput, and keeps the reference mode. H3 Max Turbo gives the same three resolutions from text or a first frame at half the H3 Max price per second, with no reference mode. All three take 5-15 seconds at 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16.
Reference-to-video with 15 files
Hailuo 03 and H3 Max accept up to 9 reference images, 3 reference videos, and 3 audio clips in one request; each reference clip runs 2-15 seconds. Use images to lock a face or a product, video for a motion or camera move to follow, and audio for a voice or rhythm the performance should sit on.
Models trained on your footage
Upload 1-40 clips of a look, a subject, or a first-to-last transition, give the model a name and a trigger phrase, and Genfire trains it in the background — usually 20-60 minutes. Attach up to three trained models to a Hailuo 03 generation, each with its own strength dial. No other model on the platform loads them.
H3 Max Director: a live session, not a render
At /dashboard/create/video/director the same model generates video faster than real time and takes new directions mid-scene — type a line and it lands in seconds, with characters, set, and story carrying across every cut. Sessions run at 480p, 768p, or 1080p in 16:9, 9:16, or 1:1, are metered by the second while live, and can be broadcast to a public Genfire Live channel viewers join in the browser.
Hailuo 03 at a glance
Prompt ideas for Hailuo 03
Built to Hailuo 03’s strengths — steal these to get started.
How to use Hailuo 03 on Genfire
- 1
Describe it
Write a prompt — or upload an image or reference media, and pick Hailuo 03 in the model selector.
- 2
Generate & compare
Render the video, iterate on the prompt, or run the same brief on a sibling model to compare — all from one credit balance.
- 3
Finish & publish
Send the result into Genfire's editor, upscaler, and audio tools, then export platform-ready output.
How the Hailuo 03 family fits together
Hailuo 03 is the top of the line and the one with the most inputs. It renders 5 to 15 seconds — any whole second, not the fixed pair of lengths Hailuo 2.3 offers — at 768p or 2K, from three kinds of brief: a text prompt, a first frame with an optional last frame to arrive at, or a reference set of up to 9 images, 3 clips, and 3 audio takes, each clip between 2 and 15 seconds. Text and reference requests pick from six aspect ratios, 21:9 through 9:16; image-to-video has no ratio control because the output follows the first frame. The end frame lives on that image-to-video path, so Genfire only offers it once a start frame is attached — otherwise the keyframe would be silently dropped from a generation you had already paid for.
H3 Max and H3 Max Turbo are fal's post-trained variants of the same model. H3 Max gives up 2K for 480p, 768p, and 1080p — 1080p bills twice the 768p rate — and in exchange follows prompts more closely and turns clips around faster. It keeps the full reference mode and the first-to-last keyframe, and accepts far longer prompts than Hailuo 03. H3 Max Turbo is the drafting tier: the same three resolutions from text or a first frame at half the H3 Max price per second, but with no reference endpoint. The practical split is Turbo to find the shot, H3 Max to finish it at 1080p, and Hailuo 03 when you need 2K or a trained model.
Trained models are what set Hailuo 03 apart from every other video model on Genfire. Under Assets you upload 1 to 40 clips — ten or more is recommended — choose what you are teaching (a Style: a look, grade, or motion language; a Subject: a person, product, or character, with up to four reference images per clip; or a Keyframe: how to move between two frames), set a short trigger phrase, and start training. It bills by training step, runs for roughly 20 to 60 minutes, and the finished model appears in the studio's Trained models picker, the References panel, the workflow editor, the API, and the MCP server. Attach up to three at once, each with a strength dial; the trigger phrase is added to your prompt automatically, and Subject models can be warm-started with more clips later. Because the adapters are trained against Hailuo 03's own weights, switching to H3 Max or Turbo removes them — the studio says so rather than ignoring them.
Loved by creators
“The interface is so intuitive. I went from signup to my first viral video in under 15 minutes.”
“I use GENFIRE for storyboarding and animatics. It communicates my vision perfectly to clients.”
“We scaled our TikTok output from 3 to 15 videos a week without hiring more editors.”
Hailuo 03 — frequently asked questions
What is Hailuo 03?
Hailuo 03 is MiniMax's third-generation video model. On Genfire it renders 5-15 second clips at 768p or 2K from a text prompt, a first frame with an optional last frame, or a set of up to 9 reference images, 3 reference videos, and 3 audio clips, and it is the only model that can run models trained on your own footage.
What is the difference between Hailuo 03, H3 Max, and H3 Max Turbo?
Hailuo 03 reaches 2K and runs trained models and references. H3 Max is a post-trained variant with stronger prompt adherence and higher throughput at 480p, 768p, or 1080p — 1080p bills twice the 768p rate — and it also takes references. H3 Max Turbo produces the same 480p, 768p, and 1080p output from text or a first frame at half the H3 Max price per second, but has no reference mode.
How many references can Hailuo 03 take?
Up to 9 reference images, 3 reference videos, and 3 reference audio clips in a single request, on both Hailuo 03 and H3 Max. Each reference clip must be between 2 and 15 seconds long. H3 Max Turbo does not have a reference mode.
What are trained models on Genfire?
A trained model is a small adapter Genfire trains on your own clips — a look, a subject, or a first-to-last transition — and applies to Hailuo 03 generations. You upload 1-40 clips, name the model and its trigger phrase, and training runs in the background for about 20-60 minutes, billed by training step. You can attach up to three trained models to one generation, each with its own strength. They run only on Hailuo 03.
What is H3 Max Director?
H3 Max Director is a live, real-time session rather than a queued render. The model generates video faster than real time, you type new directions while it plays and they land in seconds, and characters, set, and story carry across every cut. It runs at 480p, 768p, or 1080p in 16:9, 9:16, or 1:1, is billed by the second while the session is live, and can be broadcast to a Genfire Live channel that viewers watch in their browser.
How long can a Hailuo 03 video be?
Any whole number of seconds from 5 to 15, on Hailuo 03, H3 Max, and H3 Max Turbo alike. Director sessions are different: they run continuously for as long as you direct them, with a fixed cap per session and longer runs chained across sessions by handing the last frame forward.
How do I use Hailuo 03 on Genfire?
Pick Hailuo 03, H3 Max, or H3 Max Turbo in the browser video studio and type a prompt, attach a first frame, or add references. The same tiers are available through the Genfire API as video.hailuo_03, video.hailuo_03_max, and video.hailuo_03_max_turbo, and through the MCP server and CLI. Genfire is pay-as-you-go: credit packs start at $19 and every purchase includes a commercial license.
Compare with other models
From the blog
All posts →Explore Genfire
AI Tools
AI Models
Generate with Hailuo 03 today
Hailuo 03 plus 65+ other AI models, one credit balance, and a full editing pipeline — free plan included, no credit card required.
Start Free