LTX retake vs Omni Flash Edit
LTX 2.3 Retake is 10 credits. Omni Flash Edit is 13. Both edit a take you already have. Neither is a generate row.
LTX 2.3 Retake is 10 credits. Gemini Omni Flash Edit is 13 credits. Both are edit-video. Both want a take you already have. Neither is a generate. The pick is which edit contract matches the change, not which brand you generated with.
LTX 2.3 Retake Video retakes a segment of an existing clip. Source video_url plus a prompt describing the change (catalog example: “change flower to red rose”), plus optional start_time, duration, and retake_mode (for example replace_audio_and_video). Category edit-video. 10 credits. Durations 2s through 20s in even steps. audio is null. Image not required because the source is video.
Gemini Omni Flash Edit is conversational video editing — video-to-video. Category edit-video. 13 credits. Audio false. No duration enum, no output resolution listed. Image not required because the source is video_url. Google’s 27 August 2026 Omni 1.1 Flash note (Build with Gemini Omni 1.1 Flash) is a Google-surface update: 10s prior context, 10s increments to 40s cumulative, first-last interpolation, 360p drafts. That is not a third Versely slug. Do not invent /models/gemini-omni-1-1-flash. Your confirm button is still Flash Edit at 13 credits, or Gemini Omni Video if you actually needed a generate.
Same job family. Different row. A new scene from a paragraph is a generate on both brands.
Same family, different contract
Edit-video means the pixels you did not name are supposed to survive. Generate means a new sample. Teams re-roll because the sticker looks similar to a generate on the same brand. It is not similar. You now have two almosts, and the approved take is gone.
LTX Retake is a segment tool. You point at a range. You say what to replace. The rest of the clip is the reason you are here. Optional start_time and duration exist so you do not resample the whole file to fix two seconds. The even-step duration list (2, 4, 6, … 20) is the length of the retake, not a generate ladder you pick for fun. A 3-second fix buys 4s. A 21-second rewrite is not this card.
Omni Flash Edit is a conversational retouch of the take. You write the note: warmer grade, red jacket, mug swapped for the branded can. There is no 2s chip because the contract is “this file, this instruction,” not a timed insert with a start_time field on the published Versely card. Audio is off on that card. Do not brief a new line of dialogue and expect a mix.
Pick Retake when the problem is local: one object, one beat, a range you can name. Pick Flash Edit when the problem is a note on the whole take: look, wardrobe, a swap that is not “seconds 4–6.” Pick neither when the idea was wrong. Then generate.
The agent phrasing that means this is edit or redo a previous generation. Redo is allowed when the idea was wrong. Edit is required when the idea was right.
AI video editor is the production door. AI video generator will also list LTX and Omni. That does not make these two slugs text-to-video samples. No source clip, no edit.
Ten credits and thirteen credits are edit stickers
Do not compare 10 and 13 to LTX Fast at 4 or Omni Video at 9 and call edit “more expensive generating.” Fast and Omni Video invent. Retake and Flash Edit revise. A 4-credit generate that loses the approved hand is more expensive than a 10-credit segment replace that keeps it.
LTX Retake’s duration chips are even seconds to 20. Use the smallest step that covers the range. Retaking 20s because you could not find start_time is how 10 credits becomes a whole-clip rewrite you could have done as a generate. The prompt is an edit instruction, not a continuation. “Then she walks off screen” is an extend sentence. “Remove the man” / “change flower to red rose” is a retake sentence.
Omni Flash Edit has an empty duration list on the published card. You are not choosing 5s versus 10s. You are sending the file and the note. Google’s 1.1 Flash surface talks about 10s input for edit/extend and 3–10s output on Google’s door. On Versely, quote 13 credits, audio false, edit-video. Do not paste Google USD ($0.10/s appeared in older Omni Flash coverage) as a Versely price.
Audio: Retake’s field is null; the catalog mentions replace_audio_and_video as an example mode. That is a mode flag, not a native-audio generate. Flash Edit’s field is false. If the take’s mix matters, plan to restore it in the editor. Native audio is a generate feature on other rows.
Resolution: neither published edit card lists a 4K ceiling the way a generate might. Output follows the take. A 720p source will not become a 4K master because you retouched the jacket. Upscale is a different family.
Do not generate because edit felt like a second app
The LTX habit: the flower is wrong, so you re-run LTX 2.3 Text to Video Fast at 4 credits and hope the flower is red this time. You get a new kitchen. The 10-credit retake was the row.
The Omni habit: the jacket is wrong, so you re-run Omni Video at 9 credits. You get a new person. The 13-credit edit was the row.
Brand matching is optional, not a law. You can Flash-Edit an LTX take. You can Retake an Omni take. The source is a video_url. The vendor of the first generate does not own the file. Matching brands helps identity continuity sometimes. It is not required. What is required is the edit contract: segment versus conversational retouch.
If you needed a look change on the whole clip — anime, grade, restyle — Lucy Restyle is 1 credit, 720p, edit-video. That is a third row in the same family. It is not a retake of two seconds and it is not Omni’s conversational swap.
If you needed the mouth to follow a new WAV, that is lipsync, not edit-video. LTX 2.3 Audio to Video is 20 credits at 1080p. Flash Edit with audio false will not lipsync.
After either edit: captions if it ships. Silent autoplay does not care that you kept the take. Add captions to video. Captions for silent autoplay. Photoreal people still need Article 50 labels when you are in scope. An edited deepfake is still a deepfake. Article 50 disclosure.
Name the change, then the row
Write the change in one sentence. If the sentence contains a timecode, Retake. If the sentence is a note on the file, Flash Edit. If the sentence is a new scene, generate. If the sentence is a new line of dialogue, lipsync or a native-audio generate, not these two.
Confirm from the catalog. If the sticker is 4 or 9 and the input is a prompt with no video_url, you are generating. Back out. The approved take is the asset. Ten or thirteen credits is the cheap way to keep it.
FAQ
Can I use Omni Flash Edit to write a new shot from text?
No. Category is edit-video. Conversational video editing wants a source clip. A new shot is Gemini Omni Video at 9 credits, or another generate.
Is LTX Retake the same as LTX Fast?
No. Fast is text-to-video at 4 credits. Retake is edit-video at 10 credits with even-step durations to 20s and a source video. Same brand. Opposite job.
Which edit should I use if I generated on the other brand?
Either, if the file is the input. Retake if the fix is a named range. Flash Edit if the fix is a note on the take. Brand matching is optional. The contract is not.