Gemini Omni 1.1 Flash API on WaveSpeedAI: 360p to 4K, Video Edit and Reference-to-Video

Google’s Gemini Omni 1.1 Flash is available on WaveSpeedAI as four endpoints, live since 2026-08-28: google/gemini-omni-1.1-flash/text-to-video, google/gemini-omni-1.1-flash/image-to-video, google/gemini-omni-1.1-flash/reference-to-video and google/gemini-omni-1.1-flash/video-edit. All four generate 3 to 10 second clips with synchronized audio, and all four take a resolution parameter with four tiers. Pricing is per second and scales with resolution: $0.03 at 360p, $0.10 at 720p, $0.15 at 1080p and $0.30 at 4K. The rest of this post covers what changed from the earlier Omni Flash endpoints, the parameter tables, the exact price grid, and how the edit and reference modes accept input.

What Google shipped in 1.1

Google announced Gemini Omni 1.1 Flash on 2026-08-27 in a developer blog post. The announcement lists four capabilities: scene extension in 10-second increments up to a cumulative 40 seconds, first-and-last-frame specification, a 360p draft mode that Google says is up to 60% faster and a third of the cost of 720p, and 1080p and 4K output. It also describes video references of up to three seconds as a multimodal input. The model is exposed through Google AI Studio, the Gemini API and the Gemini Enterprise Agent Platform.

Two of those map directly onto parameters on WaveSpeedAI (first/last frame via last_image, video references via reference_videos), and the 360p through 4K tiers map onto resolution. Scene extension beyond a single 10-second generation is not exposed as a parameter on any of the four endpoints, so treat each request as a standalone clip.

What changed from Gemini Omni Flash

The original Omni Flash endpoints (google/gemini-omni-flash/*) are still listed but their model pages now point to the 1.1 successors. The differences that affect integration code:

	Omni Flash (gemini-omni-flash)	Omni 1.1 Flash (gemini-omni-1.1-flash)
Resolution control	none (single output size)	resolution: 360p, 720p, 1080p, 4k
Image-to-video end frame	not available	optional last_image
Reference inputs	images only	images (up to 10) plus reference_videos (up to 3, each 3 seconds or less)
Duration	3 to 10 s, default 8	3 to 10 s, default 8
Aspect ratio	16:9, 9:16	16:9, 9:16
Price	flat $0.13 per second	$0.03 to $0.30 per second by resolution

The pricing change cuts both ways. At 720p, the default, an 8-second clip drops from $1.04 to $0.80. At 1080p the same clip costs $1.20, and at 4K $2.40, both higher than the old flat rate. If you were running Omni Flash for final delivery and want the higher tiers, budget for it; if you only need 720p or draft output, 1.1 is cheaper.

Parameters

Text-to-video and image-to-video share the core set. Reference-to-video and video-edit differ in their input fields.

Parameter	text-to-video	image-to-video	reference-to-video	video-edit
prompt (required)	yes	yes	yes	yes (edit instruction)
image	-	required start frame URL	-	-
last_image	-	optional end frame URL	-	-
images	-	-	up to 10 reference image URLs	-
reference_videos	-	-	up to 3 URLs, each 3 s or shorter	-
video	-	-	-	required source URL, 10 s max
aspect_ratio	16:9 (default), 9:16	same	same	not applicable
resolution	360p, 720p (default), 1080p, 4k	same	same	same
duration	integer 3 to 10, default 8	same	same	derived from source

Reference-to-video requires at least one of images or reference_videos alongside the prompt. Video-edit has no duration or aspect_ratio field; the output follows the source clip.

Pricing

The live formula for the three generation endpoints is base_price * duration * multiplier, with a base price of $0.10 per second and multipliers of 0.3 (360p), 1 (720p), 1.5 (1080p) and 3 (4K). Evaluated for the common durations:

Resolution	Per second	3 s	5 s	8 s (default)	10 s
360p	$0.03	$0.09	$0.15	$0.24	$0.30
720p	$0.10	$0.30	$0.50	$0.80	$1.00
1080p	$0.15	$0.45	$0.75	$1.20	$1.50
4K	$0.30	$0.90	$1.50	$2.40	$3.00

Video-edit uses the same per-second rates but the billed duration comes from the source file: it is rounded up to the next whole second, then clamped to a minimum of 3 and a maximum of 10. A 4.2-second source is billed as 5 seconds ($0.50 at 720p, $1.50 at 4K); a 2-second source is billed as 3 seconds; a 12-second source is rejected by the 10-second input limit rather than billed at 10.

The practical pattern is the one Google’s own announcement suggests: iterate at 360p, where a full 10-second clip is $0.30, then re-run the winning prompt at 1080p or 4K. Ten 360p drafts plus one 4K final for an 8-second clip is $2.40 + $2.40 = $4.80, versus $26.40 if every iteration ran at 4K.

How image-to-video with an end frame works

image sets the first frame. When last_image is also supplied, the model generates the motion between the two frames, which is the first/last-frame control from Google’s announcement. Use it for transitions, reveals and camera moves where the end composition matters; leave it out when you only want the start image animated. The prompt still carries the motion, camera and audio description in both cases.

{
  "image": "https://example.com/product-front.png",
  "last_image": "https://example.com/product-side.png",
  "prompt": "Slow orbit from front to side view, studio lighting, soft ambient music",
  "resolution": "1080p",
  "duration": 5,
  "aspect_ratio": "16:9"
}

How reference-to-video works

Reference-to-video takes a prompt plus reference media. images (up to 10) guide subject identity, product appearance or environment; reference_videos (up to 3, each 3 seconds or shorter) guide motion, gesture or camera behavior that is hard to describe in text. The model page recommends stating in the prompt what each reference should influence, and not stacking conflicting references in one request. This is the endpoint to use for character-consistent shots across a series, or for transferring a specific movement from a short clip onto a new subject.

How video-edit works

Video-edit takes a video URL of up to 10 seconds and a natural-language prompt describing the change: restyle, swap a subject’s outfit, alter lighting, adjust the background. The output resolution is chosen independently of the source, so you can preview an edit direction at 360p and render the final at 1080p or 4K. The WaveSpeedAI schema exposes no audio-reference or voice-editing input on this endpoint, so edit prompts should describe visual changes.

Calling the endpoints

Every endpoint follows the same request pattern: POST https://api.wavespeed.ai/api/v3/<model_uuid> with a JSON body, then poll GET /api/v3/predictions/<id>/result until the status is completed. For example:

curl -X POST https://api.wavespeed.ai/api/v3/google/gemini-omni-1.1-flash/text-to-video \
  -H "Authorization: Bearer $WAVESPEED_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "A ceramic mug on a wooden table, morning light, steam rising, quiet kitchen ambience", "resolution": "360p", "duration": 5, "aspect_ratio": "9:16"}'

Model pages, with schema, playground and live pricing:

• Text-to-video
• Image-to-video
• Reference-to-video
• Video edit

Related reading

• Gemini Omni Flash shipped: what actually launched covers the original May release and its 10-second design decision.
• Omni Flash pricing breaks down Google’s consumer credit pricing, which is a different cost model from the per-second API rate above.
• Gemini Omni Flash vs Seedance 2.0 vs Kling 3.0 compares the multimodal video models available on the platform.