What is MiniMax Hailuo 02?
MiniMax Hailuo 02 is a generative video model released by MiniMax on June 18, 2025. It was built for short-form video creation rather than general conversation, coding, image synthesis, audio generation, or autonomous task execution. The model accepts written descriptions and image-based prompts, then produces short video clips that interpret the requested subjects, actions, camera movement, and scene changes.
At launch, Hailuo 02 was available through the Hailuo Video consumer platform, its mobile application, and the MiniMax developer API. MiniMax positioned it as an improvement for scenes where movement and physical interactions are difficult for video models to render consistently. Examples include dynamic camera shots, objects changing position, characters performing actions, and scenes that require a clear sense of motion or transformation.
The model is no longer prominent in MiniMax's current first-party catalog. MiniMax later introduced Hailuo 2.3 in October 2025, and current documentation emphasizes newer H3 video models. There is no supplied evidence of a formal shutdown date for the exact Hailuo 02 identity, but its current status is best described as legacy or superseded.
Core video-generation capabilities
Hailuo 02 supports two primary workflows:
- Text-to-video: a written prompt describes the scene, subject, action, style, and camera behavior.
- Image-to-video: a supplied still image provides a visual starting point that the model animates into a short clip.
Its documented launch configurations included 768p video at six or ten seconds and 1080p video at six seconds. The model was also associated with 512p six- and ten-second configurations in historical API descriptions. These are short clips, so Hailuo 02 is intended to generate individual shots that can be selected, edited, and combined rather than complete long-form productions in one request.
MiniMax highlighted several areas of improvement in its launch materials:
- Native 1080p video output rather than relying only on later upscaling.
- More detailed instruction following for prompts containing several actions or visual requirements.
- More convincing physical movement and interactions between objects or characters.
- Cinematic camera movement for short advertising, social, and concept-video shots.
- Improved inference efficiency, which MiniMax associated with lower generation costs.
These capabilities describe the provider's positioning and launch claims. They should not be read as a guarantee that every prompt will produce consistent motion, accurate physics, or repeatable characters.
Start-and-end-frame control
In August 2025, MiniMax added start-and-end-frame functionality for Hailuo 02. This allows a creator to specify the visual state at the beginning and at the end of a generated clip. The model then attempts to create a transition between those states.
The update supported start-and-end-frame workflows at 768p and 1080p. Start-frame functionality was also supported at 512p. This type of control is useful when the desired result is not simply “animate this image,” but rather “move from this defined opening composition to that defined closing composition.”
Practical uses include product advertisements that transition between two product positions, cinematic transformations, visual effects tests, and shots where the opening and closing compositions need to be planned in advance. Frame control does not remove the need for experimentation: the model still has to invent the intermediate motion, so complex transitions may produce unexpected object changes or inconsistent details.
Supported inputs and outputs
| Capability | Hailuo 02 support |
|---|---|
| Text input | Yes |
| Image input | Yes, for image-to-video generation |
| Video input | Not documented in the supplied specifications |
| Video output | Yes |
| Text output | No |
| Image output | No |
| Audio, speech, or music output | No |
| Embedding output | No |
The model's “multimodal” classification refers to its ability to work with text and images as inputs while producing video. It should not be treated as a general multimodal assistant. Hailuo 02 is not documented as a language model with a context window, text completion limit, reasoning mode, coding ability, or knowledge cutoff. Context-length and maximum-token fields are therefore not applicable to its primary video-generation role.
Pricing and availability
Historical Hailuo 02 API pricing was configured per generated video rather than as an input-token and output-token rate. The documented configurations included 512p/6-second, 512p/10-second, 768p/6-second, 768p/10-second, and 1080p/6-second outputs. The supplied research does not verify a current first-party price for the original Hailuo 02 model, so a precise present-day cost should not be assumed.
Hailuo 02 was offered through MiniMax's consumer-facing Hailuo products as well as the MiniMax API platform. Availability, quotas, and access could vary by product and region. Because the current MiniMax catalog primarily highlights newer video systems, users evaluating the model today should first confirm that Hailuo 02 is still selectable in the relevant interface or API account.
Its historical efficiency emphasis suggests a favorable cost-versus-quality position for short clips, but the supplied material does not provide a verified benchmark or current price comparison. Any cost assessment should therefore be based on the actual configuration and account pricing currently shown by MiniMax.
Strengths and limitations
Where Hailuo 02 was strongest
- Short cinematic shots: its design suits individual scenes for trailers, social posts, storyboards, and visual experiments.
- Motion-heavy prompts: it was specifically positioned for complex physical actions and dynamic scenes.
- Higher-resolution delivery: native 1080p was available for six-second generation configurations.
- Prompt-driven control: detailed descriptions could specify actions, camera behavior, and scene changes.
- Planned transitions: start-and-end-frame tools helped define the intended trajectory of a shot.
- Multiple access routes: creators could use consumer Hailuo products, while developers could access the model through MiniMax's API when available.
Important limitations
- Short output duration: the documented outputs were six or ten seconds, which makes editing or shot assembly necessary for longer videos.
- No native sound: Hailuo 02 did not generate audio, speech, music, or sound effects.
- No general-purpose reasoning: it was not a chatbot, coding model, tool-using agent, or analytical language model.
- No verified current price: historical configuration pricing exists, but the exact current first-party price for the original model identity could not be confirmed.
- Legacy status: newer MiniMax models now receive greater emphasis in current documentation, which may affect access, support, and future availability.
- Generative inconsistency: as with other video generators, complex scenes may not preserve every detail or produce physically perfect intermediate motion. This is an evaluation consideration rather than a provider-published guarantee.
Reasoning, coding, and tool support
Hailuo 02 should not be evaluated using the same criteria as a text-based reasoning or coding model. It does not produce text answers as its primary output, has no documented reasoning mode, and is not intended to write or execute code. It also has no documented function-calling, web-search, agent, or tool-use capability.
Its “instruction following” refers to interpreting a visual-generation prompt, not to multi-step reasoning in the language-model sense. A prompt can describe several actions or camera instructions, but the result is still a generated video rather than an explanation, plan, or executable workflow.
When to choose Hailuo 02
Hailuo 02 makes the most sense when a compatible MiniMax interface still exposes it and the task calls for a short visual shot rather than a complete production pipeline. It can be a reasonable choice for:
- Advertising concepts and product demonstrations.
- Social-media clips and short-form promotional content.
- Character or object motion tests.
- Storyboard and previsualization work.
- Image animation with a specified opening or closing state.
- Cinematic experiments involving camera movement or complex physical action.
Choose a newer MiniMax video model instead when current availability, ongoing support, improved character motion, stronger stylization, facial expression quality, or newer motion controls matter more than reproducing an older workflow. MiniMax's Hailuo 2.3 was announced as a successor that built on Hailuo 02, while current H3 documentation represents a newer generation of MiniMax video offerings.
A different type of tool is more appropriate when the project needs long-form continuity, synchronized sound, speech or music generation, extensive editing, or a general-purpose assistant that can plan and execute production tasks. Hailuo 02 is best treated as a short-shot generator, not as an all-in-one video studio.
Bottom line
MiniMax Hailuo 02 was an important 2025 MiniMax video model focused on short text-to-video and image-to-video generation. Its most useful distinctions were native 1080p output, attention to complex motion, and start-and-end-frame controls. Those features made it suitable for cinematic snippets, advertisements, social content, and visual prototyping.
Its present-day drawback is positioning: Hailuo 02 is no longer the clearest current entry point in MiniMax's video catalog, and its exact current pricing and availability are not verified in the supplied documentation. For a new project, confirm access before designing a workflow around it and compare the result with MiniMax's newer video models. For historical evaluation or a compatible account that still exposes the model, Hailuo 02 remains a focused option for generating short, visually directed clips.