Head to head: Gemini Omni Flash 1.1 Image to Video vs LTX-2.3 22B Distilled
Gemini Omni Flash 1.1 Image to Video vs LTX-2.3 22B Distilled
By Ryan Merket · Published · Updated
A comparison of two image-to-video models across lighting changes, rooftop choreography, roller-skating follow shots, and fluid effects. The verdict weighs prompt fidelity and how coherently each model carries action and visual changes across frames.
Gemini Omni Flash 1.1 Image to Video leads the saved task sample with an aggregate score of **32.5 to 15.8** and **12 task wins to 0**, with no ties. The result was consistent across three successful trials of each of the four briefs—not a close split between different strengths. Gemini’s edge was especially clear in prompt fidelity. It kept the rooftop dancer, rust-red skirt, steaming kettle, and laundry in the requested setting; for the roller follow, it preserved the wet riverfront alley, distinctive outfit, parasol, and front-facing tracking. LTX-2.3 22B Distilled often offered attractive imagery, but shifted those scenes to substantially different settings or compositions. The same pattern held in the lighting and fluid briefs. Gemini delivered a more legible warm-to-cool dusk transition with the lamp coming on, and more clearly showed a balloon becoming a spreading sheet of water and scattered droplets. LTX’s alternatives were less faithful to the requested transitions and burst dynamics. **Verdict: Gemini Omni Flash 1.1 Image to Video is the clear winner on this task sample.** That assessment is limited to these saved tasks; it should not be read as a claim about either model’s general capabilities.
Lighting transition
A living room at dusk as the sun sets over a few seconds: the warm light visibly dims and shifts to cool blue while a lamp flickers on, the transition smooth and believable, locked-off frame, 16:9.
OpenAI: GPT-6 Luna: Model A shows a convincing, smoothly progressing sunset shift from warm light to cool dusk, with a lamp coming on in a stable, locked-off composition. Model B changes from day to night but adds a moving character and has a more abrupt, less faithful lighting transition. (Second judge pass, order swapped — scores are the average of both: Model A shows a believable sunset-to-blue-dusk transition in a locked-off living room, with the lamp switching on by the final frame. Model B shifts from warm da | Anthropic: Claude Fable 5.1: Model A delivers a photoreal locked-off living…
Rooftop kettle dance
One continuous 8-second shot, 16:9: on a cramped apartment rooftop garden at first light, a barefoot contemporary dancer in a rust-red rehearsal skirt performs a slow turning phrase beside a dented brass kettle on a camping stove, lifting her arms through curls of steam while laundry lines flutter overhead; the camera makes a gentle sideways dolly from left to right with a slight handheld sway, maintaining a medium-wide profile view of her full body as the environment stays alive around her—pale clouds drifting behind water towers, basil leaves trembling in the breeze, steam rising continuously from the kettle, and a shallow tray catching ripples from occasional drips; warm pink sunrise light, hushed reflective mood, natural continuous motion throughout.
OpenAI: GPT-6 Luna: Model A closely matches the requested rooftop garden, barefoot dancer, rust-red skirt, steaming kettle, laundry, and gentle profile framing, with a graceful evolving phrase. Model B has appealing sunrise imagery and arm movement, but shifts to a much larger seaside setting and includes an additional distant dancer. (Second judge pass, order swapped — scores are the average of both: Model A better matches the rooftop garden, steaming brass kettle on a camping stove, laundry, barefoot dancer, and | Anthropic: Claude Fable 5.1: Model A closely matches the prompt with a barefo…
Neon roller follow
One continuous 7-second shot, 16:9: at blue-hour in a nearly empty riverfront market alley, a woman in a silver fringe jacket and bright mango roller skates glides fast while twirling a folded orange parasol like a dance prop, weaving past stacked mint-green crates and wet cobblestones; the camera jogs backward in an energetic handheld documentary style, smoothly tracking her from the front and subtly rising and dipping to keep her centered from knees to head as she accelerates, while reflections from flickering violet shop signs streak across the puddles behind her; soft drizzle, cool neon lighting, playful triumphant mood, sharp focus locked on her face and spinning parasol as the background rushes by.
OpenAI: GPT-6 Luna: Model A closely matches the riverfront alley, wet cobblestones, silver jacket, orange parasol, yellow skates, and front-facing tracking, with consistent movement across the frames. Model B has energetic neon visuals and crates, but the indoor tunnel and substantially different outfit depart from the prompt. (Second judge pass, order swapped — scores are the average of both: Model A much more closely matches the wet market alley, silver jacket, bright skates, crates, and front-tracking feel, with | Anthropic: Claude Fable 5.1: Model A nails the blue-hour riverfront alley, w…
Fluid & particle dynamics
Cinematic slow-motion of a water balloon bursting, the sheet of water expanding and droplets scattering through the air in convincing detail against a dark background, hard side light, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly shows the progression from a rupturing balloon to an expanding water sheet and scattered droplets, with convincing hard-lit detail against black. Model B has a strong dark cinematic look, but its water appears more like bubble-like beads and thin filaments than a broad bursting sheet. (Second judge pass, order swapped — scores are the average of both: Model A clearly shows the balloon rupturing and transitioning into a spreading water membrane and scattered droplets, with co | Anthropic: Claude Fable 5.1: Model A shows a plausible sequence—balloon skin…
Matchup powered by OpenRouter.