Luma
Luma Ray 2 is the specialist for camera-driven image animation on Martini. While Kling 3.0 excels at moving subjects (people, objects), Ray 2 excels at moving the camera — producing smooth dollies, orbits, zooms, and pans that feel like a real cinematographer's work rather than digital effects. It also adds a distinctive filmic quality (natural grain, cinematic color grading) that other models don't replicate. Ray 2 offers three resolution tiers (540p, 720p, 1080p) for 5-second clips, with output detail and render time scaling up at each step. The lighter Ray Flash 2 variant generates at 540p faster than full Ray 2 — useful for testing camera angles before committing to a high-resolution final render.
The most important decision in image animation isn't which model to use — it's whether your scene needs camera motion or subject motion. Ray 2 excels when the image itself is relatively static but you want the viewer to feel like they're moving through the scene: a landscape where the camera drifts forward revealing depth, a product where the camera orbits to show dimension, or an interior where the camera pans across architectural detail. If you need the subjects to move (a person blinking, waves crashing, leaves falling), Kling 3.0 or Seedance 2.0 will produce better results. Many beginners waste render time trying to make Ray 2 animate subject motion — it can do it, but that's not its strength.
Ray 2 interprets prompts as camera operator instructions, not scene descriptions. Use cinematographic vocabulary: "Slow dolly forward" (camera moves toward the subject on a track), "gentle orbit left to right, 180-degree arc" (camera circles the subject), "crane shot rising upward" (camera lifts vertically). The more specific your camera instruction, the more predictable the result. Compare: "move closer to the building" (vague — could be a zoom, dolly, or crop) vs "slow dolly forward, keeping the building centered, foreground elements passing by on both sides" (precise — the model knows exactly what parallax to create). Ray 2 infers depth from your 2D image and creates natural parallax as the virtual camera moves, so detailed camera direction directly improves the 3D effect.
Different camera movements serve different purposes, and choosing the wrong one wastes a render on an underwhelming result. Dolly forward/backward works best for landscapes and environments — it reveals depth layers and creates an immersive "walking into the scene" feeling. Orbit works best for single subjects and products — it shows dimension and reveals angles the static image can't. Pan works best for wide scenes and panoramas — it simulates turning your head across a vista. Crane/vertical works best for architectural and dramatic reveals — it creates a sense of scale. Zoom (as opposed to dolly) compresses depth rather than revealing it, making it ideal for dramatic focus pulls. The key insight: dolly and zoom look completely different even though beginners often conflate them. A dolly creates parallax (foreground moves faster than background); a zoom does not.
Ray Flash 2 generates at 540p faster than full Ray 2 at the same resolution. Use Flash to test 3-4 different camera movements and find the one that works best with your image. Once you identify the winning angle and speed, switch to full Ray 2 for the final render. The camera movement physics are identical between Flash and full Ray 2 — Flash just generates faster at slightly lower visual detail. For a typical project: 4 Flash tests plus 1 final Ray 2 at 720p completes much faster than iterating every pass in Ray 2 720p directly. The Flash-first workflow keeps exploration noticeably faster.
Dolly forward — Ray 2's signature move. This prompt works because it gives three layers of instruction: the primary camera motion ("dolly forward"), the desired depth effect ("natural parallax between foreground and background"), and an atmospheric cue ("haze catching light") that adds volumetric depth. Ray 2 separates your flat image into depth layers and moves through them, creating a 3D parallax effect from a 2D photo. The atmospheric haze cue encourages the model to add subtle light scattering between layers, enhancing the sense of distance.
Slow cinematic dolly forward, revealing depth in the scene. Natural parallax between foreground and background elements. Subtle atmospheric haze catching light.
Orbital camera — perfect for product or portrait animation. The "180-degree arc" gives a precise angular instruction that Ray 2 interprets literally: the camera will trace exactly half a circle, not a vague "orbit." "Shallow depth of field" keeps the subject sharp while the background blurs during the orbit, mimicking real lens behavior. The "cinematic film grain" reinforces Ray 2's natural filmic rendering — the model already adds subtle grain, and this cue amplifies it for a more analog, less digital aesthetic.
Gentle orbital camera movement around the subject, 180-degree arc from left to right. Shallow depth of field maintains focus on the center. Cinematic film grain.
Ray 2 render time scales with resolution across 540p, 720p, and 1080p tiers. For social media (Instagram, TikTok), 720p is sufficient and renders noticeably faster than 1080p. Reserve 1080p for portfolio pieces, client deliverables, or large-screen presentations where the extra detail is visible.
Specify exact camera arcs and directions: "90-degree orbit" is better than "orbit around." "Slow dolly forward with foreground elements passing by" is better than "move closer." Ray 2 interprets camera instructions literally, so precision equals predictability.
Ray 2 adds natural film grain and cinematic color grading automatically — it's the only model on Martini with this built-in aesthetic. If you want a cleaner, more digital look, Kling 3.0 or Seedance 2.0 will produce sharper, grain-free results. This isn't a quality difference — it's an aesthetic choice.
For product photography animation (turntable-style reveals), Ray 2's orbital camera at 720p produces the most professional-looking results. Kling 3.0 Standard is the lighter alternative for the same shot — faster but less cinematic.
Ray 2 produces the most cinematic-looking image animations on Martini, with a distinctive filmic quality that no other model replicates. The natural grain, warm color grading, and smooth camera physics feel like real drone or Steadicam footage, not AI-generated clips. The trade-off is clear: Ray 2 is a camera-motion specialist, not a subject-motion generalist. For animating people or dramatic subject action, Kling 3.0 Pro produces more natural human motion. For high-energy dramatic animations from illustrations, Seedance 2.0 Pro is the better choice. Ray 2's sweet spot is landscape photography, product shots, architectural visualization, and any image where sweeping camera work is the hero. At 720p, it sits at a heavier render tier than both Seedance Pro and Kling Pro — a fair trade for the unique cinematic aesthetic.
Connect Ray 2 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started FreeKling
Kling 3.0 is the best model on Martini for animating images that contain people. Its Pro tier generates the most natural facial expressions, body movement, and hair physics of any image-to-video model. Two tiers are available: Standard for fast iteration and Pro for delivery-grade output. The quality gap on human faces is dramatic — Pro handles the micro-expressions (blinking rhythm, mouth corners lifting, subtle head tilts) that make the difference between "obviously AI" and "wait, is that real?" For landscapes and objects without people, Standard is perfectly sufficient.
View guideByteDance
Seedance 2.0 by ByteDance is optimized for dramatic, high-energy image animations — the kind of dynamic action that Kling 3.0 handles competently but Seedance handles exceptionally. Capes billowing, swords swinging, particles exploding, environmental destruction: Seedance turns these into fluid, cinematic clips. The model offers three tiers: Fast for rapid motion exploration, Standard for publishable quality, and Seedance 2 Pro for maximum detail. It supports 6 aspect ratios including 21:9 ultra-widescreen, and works with both image-to-video and text-to-video. The Omni Pro variant additionally supports video-to-video and reference images for even more control.
View guide