Hailuo 2.3 pro: Image-to-Video at 1080p with Fluid, Physics-Aware Motion
hailuo 2.3 pro is MiniMax's 1080p image-to-video model. Give it one still image and a motion prompt, and it animates the scene — composition, color, and subject held steady while it adds camera movement and natural motion. On GPTProto you call the hailuo 2.3 pro API at $0.441 per generation, about 10% below MiniMax's direct rate, from one key and one USD balance shared across 200+ models. Browse all video models to compare options.
What is Hailuo 2.3 pro?
Within MiniMax's Hailuo 2.3 family, hailuo-2.3-pro is the high-resolution tier — built by MiniMax, a Shanghai-based AI foundation model company, and released October 28, 2025. It reads your input image as the opening frame and builds motion outward from it, keeping the original composition, color, and subject identity intact across the clip while adding camera movement and lighting transitions. On this page the model runs in image-to-video mode: supply one image plus a prompt describing how the scene should move.
The 2.3 generation improves on Hailuo 02 in body-movement fluidity, facial micro-expressions, physics, and prompt adherence, and supports anime, illustration, ink-wash, and game-CG styles.
Specs
| Field | Value |
|---|---|
| Provider | MiniMax (Shanghai) |
| Model string | hailuo-2.3-pro |
| Modality (this page) | Image-to-video |
| Output resolution | 1080p |
| Clip duration | 6–10s |
| Aspect ratios | Multi-aspect |
| Audio | None — no native audio track |
| Output | Single continuous shot per call |
| Styles | Photoreal, anime, illustration, ink-wash, game-CG |
| Generation mode | Asynchronous (submit task → poll → retrieve MP4 URL) |
Hailuo 2.3 pro vs Hailuo 2.3 fast
Both tiers are live on GPTProto, so this is a real same-family comparison — pick by quality vs cost, not marketing.
hailuo-2.3-pro |
hailuo-2.3-fast |
|
|---|---|---|
| Resolution | 1080p | 768p |
| Modality | Image-to-video (text-to-video on its own page) | Image-to-video only |
| Relative speed | Standard | Faster turnaround |
| Price / generation | $0.441 | $0.171 |
| Best for | Final renders, client-facing 1080p clips | Batch prototyping, e-commerce variants, concept tests |
Rule of thumb: iterate on Fast (≈61% cheaper), render the keeper on Pro. If you specifically need last-frame conditioning, the prior-gen hailuo-02-pro supports it — 2.3 does not.
Switching from MiniMax direct
If you already call MiniMax's Open Platform, moving over is a base-URL + key swap:
- Endpoint →
https://gptproto.com/api/v3/minimax/hailuo-2.3-pro/image-to-video - Auth → your
GPTPROTO_API_KEY - Model →
hailuo-2.3-pro
What changes: you pay from one USD balance instead of MiniMax video points or monthly credit packs, you skip the consumer-vs-developer plan split, and you avoid region-gated sign-up. Same model, ~10% lower per-clip cost.
Content policy & use
hailuo 2.3 pro runs through GPTProto's content review; generations that fail review aren't billed. When enable_prompt_expansion is on, the model's safety checker is active and filters risky prompts. GPTProto does not provide a way around that filtering — if a generation is blocked, adjust the prompt or input image rather than seeking a bypass. Output is usable commercially under MiniMax's model terms; you're responsible for your input images and prompts.
Hailuo 2.3 pro Prompt recipes (image-to-video)
hailuo 2.3 parses prompts spatially — open with camera logic, then motion, then resolve on stillness. Three patterns that hold up:
- Camera-first push-in —
Slow dolly-in on [subject]. [one motion]. [lighting]. Settles as [subject] comes to rest. - Product reveal —
Static product on [surface]. Slow 90° turntable. Reflective highlights track the rotation. Hold on the final angle. - Character beat —
Medium shot, [subject] with a calm, confident expression. [single gesture]. Subtle smile, then lowers gaze.
Keep to one clear motion idea per clip — 2.3 is single-shot, so multi-action prompts split focus.
