Gemini Omni 1.2 Flash is live in Picsart, and the headline change over Gemini Omni 1.1 is 4K: the model now generates and edits at 4K rather than stopping at 1080p. Everything else that makes Omni distinctive carries forward intact, so this is a resolution upgrade to a model that was already generating video and synchronized audio in a single pass.<\/p>\n
That makes the choice between the two versions unusually simple. The question is not which model has more features. The question is whether the shot needs to survive past 1080p.<\/p>\n
| Capability<\/th>\n | Gemini Omni 1.1<\/th>\n | Gemini Omni 1.2 Flash<\/th>\n<\/tr>\n |
|---|---|---|
| Generation and editing resolution<\/td>\n | Up to 1080p<\/td>\n | Up to 4K<\/td>\n<\/tr>\n |
| Reference images per generation<\/td>\n | Supported<\/td>\n | Up to five in Picsart<\/td>\n<\/tr>\n |
| Reference videos per generation<\/td>\n | Supported<\/td>\n | Up to three in Picsart<\/td>\n<\/tr>\n |
| Start and end frame<\/td>\n | Supported<\/td>\n | Supported, set independently<\/td>\n<\/tr>\n |
| Clip extension<\/td>\n | Supported<\/td>\n | Supported<\/td>\n<\/tr>\n |
| Maximum clip length<\/td>\n | 10 seconds<\/td>\n | 10 seconds<\/td>\n<\/tr>\n |
| Aspect ratios<\/td>\n | 16:9 and 9:16<\/td>\n | 16:9 and 9:16<\/td>\n<\/tr>\n |
| Video plus synchronized audio<\/td>\n | One pass<\/td>\n | One pass<\/td>\n<\/tr>\n |
| Chat-edit after generating<\/td>\n | Yes<\/td>\n | Yes<\/td>\n<\/tr>\n |
| On-screen text rendering<\/td>\n | Yes<\/td>\n | Yes<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/figure>\n The pattern in that table is the point. Version 1.2 Flash is not a rebuilt model with a new feature list. It takes a capable model and lifts the ceiling on output quality.<\/p>\n What 4K generation and editing unlocks<\/span><\/h2>\nThe footage survives a real edit.<\/strong> Color grading, stabilization, and speed ramps all cost detail. Starting at 4K leaves headroom for those passes, so the version that reaches the timeline is not already the version that has nothing left to give.<\/p>\n Editing happens at 4K too.<\/strong> The resolution lift covers the chat-edit loop, not only the first generation. Swapping an object or changing a wardrobe color returns a 4K frame, so an edited clip does not quietly drop back to a lower-resolution master partway through the process.<\/p>\n It clears the bar for work that leaves a feed.<\/strong> Anything projected, shown on a large display, cut into a broadcast package, or pulled as a still for print needs more than 1080p. Version 1.1 could not serve any of it without an upscaling pass afterward.<\/p>\n The capabilities below are not new in 1.2 Flash, and they are worth knowing because they are the reason Omni is worth generating with at 4K in the first place. Picsart sets the specific slot counts on the model card.<\/p>\n Five reference images.<\/strong> The Picsart card takes up to five reference images in one generation, so a single shot can carry a product, a person, a setting, and a look at once. There is no need to composite the ingredients into one image before prompting.<\/p>\n Three reference videos.<\/strong> Motion becomes reference material rather than something described in words. Feed the model the camera move you want alongside the performance you want, and ask for a shot that honors both.<\/p>\n A start frame and an end frame.<\/strong> Both slots are set independently and the model builds the frames between them. Bracketing a move with its first and last image is a far more predictable way to land a specific shot than describing the move.<\/p>\n Clip extension.<\/strong> Existing footage goes in and the model continues the shot instead of starting a new one. That is the piece that makes Omni useful on projects that began outside of it.<\/p>\n Any length from three to ten seconds.<\/strong> The Picsart card offers every whole second in that range, which matters when a cut needs exactly four or seven seconds and the alternative is generating longer and trimming.<\/p>\n Version 1.1 does not become obsolete, and picking it deliberately is sometimes the better call.<\/p>\n A social-only deliverable.<\/strong> A vertical cut destined for one feed gets compressed on upload regardless. Generating above the delivery spec costs more and changes nothing a viewer will see.<\/p>\n Fast iteration on an idea.<\/strong> Testing ten variations of a concept is cheaper at a lower resolution. Settle the composition first, then generate the keeper at 4K.<\/p>\n No reframing planned.<\/strong> The 4K argument leans heavily on cropping and grading headroom. A shot going out in the ratio it was generated in, ungraded, does not need that headroom.<\/p>\n The upgrade is easiest to judge on a shot you intend to crop. Generate this at 16:9 in 4K, then pull a 9:16 vertical out of the same frame and look at how the vertical holds up.<\/p>\n Slow handheld push-in on a ceramic coffee cup on a weathered wooden table, steam rising, a blurred cafe window behind it. Start wide enough to show the full table and end tight on the rim of the cup. Warm late-afternoon light raking in from the left, soft shadow falling to the right, shallow depth of field, fine grain on the wood grain and the glaze. Quiet room tone, a distant espresso machine, one soft ceramic clink as the camera settles<\/p>\n\n |