Skip to studio content
ProjectOpening projects…On this device

Drafts and selected assets stay in this browser. Export a copy before clearing site data. Imported projects never resume paid requests. Large provider-hosted media may need downloading separately before its link expires.

Project & assets ↗

Configured routes · Reviewed 10 September 2026

The model landscape, and why routing beats picking

There is no single best model. There are jobs, and models that are currently best at them, at a price. A production studio's edge is a routing table it re-evaluates monthly, not a favorite tool. This page is the live routing table behind the Studio demo.

Choose for this job

Compare the scenario, not the sticker price.

These estimates use the same configured pricing logic as the studio. They are not live vendor quotes. Input-video billing, image tiers and started-minute audio billing are included where implemented.

gemini · Supported by this app

Veo 3.1 Fast (Google)

~$1.200

8s · configured per-second rate; resolution is endpoint-controlled here

Drafting, social clips, high-volume video versioning.

Native audio · No audio-reference input

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

gemini · Supported by this app

Veo 3.1 (Google)

~$6.000

8s · configured per-second rate; resolution is endpoint-controlled here

Hero spots and finished broadcast-style deliverables.

Native audio · No audio-reference input

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

fal · Adjust this scenario

Kling 3.0 (Kuaishou, via fal.ai)

—

8s · configured per-second rate; resolution is endpoint-controlled here

  • Choose 5 or 10 seconds.

Cost-efficient social cutdowns; same product across many shots.

Silent output · No audio-reference input

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

fal · Adjust this scenario

Kling 3.0 Pro (Kuaishou, via fal.ai)

—

8s · configured per-second rate; resolution is endpoint-controlled here

  • Choose 5 or 10 seconds.

A finished-looking single shot when Veo is too expensive and Standard is not holding up.

Native audio · No audio-reference input

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

fal · Supported by this app

Runway Gen-4 (via fal.ai)

~$1.200

8s · configured per-second rate; resolution is endpoint-controlled here

VFX-leaning brand films and shot-heavy storytelling where you expect to iterate on the motion.

Silent output · No audio-reference input

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

fal · Supported by this app

H3 Max Reference (MiniMax, via fal.ai)

~$0.400

8s · 480p · output + reference inputs; assumes 1024×1024 per image; 1080p input rate estimated from 768p

Blender-guided product films with recurring subjects. Output and reference inputs are billed separately; exact edit timing still needs review.

Native audio · Audio references accepted

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

fal · Supported by this app

Seedance 2.5 (ByteDance, via fal.ai)

~$1.643

8s · 480p · 9:16 · 0s video input

A move between two frames you have already approved, and the cheapest Seedance route, since nothing is billed as input duration.

Native audio · No audio-reference input

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

fal · Supported by this app

Seedance 2.5 Reference (ByteDance, via fal.ai)

~$1.643

8s · 480p · 9:16 · 0s video input

Product-identity-critical ads: the packaging must stay exactly itself while the camera moves.

Native audio · Audio references accepted

Configuration review: 10 Sep 2026. Verify current vendor terms and rates.

Provider source ↗

The destination shows its own supported options and final estimate before any generation. Model access is verified by the provider only when a request is made.

Inspect the project’s actual inputs and results ↗
Configured model inventory
ModelBilling basisJob it owns here
Nano Banana Pro (Gemini 3 Pro Image)Output size + applicable referencesPromo tiles with live text, bilingual versioning, brand-critical hero shots.
Nano Banana (Gemini Flash Image)Output size + applicable referencesFormat adaptations, seasonal variants, bulk versioning at scale.
Flux 2 Pro (Black Forest Labs)Output size + applicable referencesPhotorealistic hero stills where the image is the product.
Flux Kontext Max (BFL, via fal.ai)Output size + applicable referencesPackshot challenger: identity-true edits where texture matters more than dense label text.
Seedream 4.0 Edit (ByteDance, via fal.ai)Output size + applicable referencesPackshot challenger: the value benchmark every bake-off should include.
GPT Image 2 Edit (OpenAI, via fal.ai)Output size + applicable referencesPackshot challenger where the label has to survive: the angle changes and the type on the pack stays readable and correct.
Recraft V4.1 Utility (1K)Output size + applicable referencesIterating: testing how a label sits on a form, or what a restage prompt actually does, before spending Utility Pro money on the final.
Recraft V4.1 Utility ProOutput size + applicable referencesThe final render of a restage you already proved on Utility 1K. Not angle synthesis. That needs a model that reads several references at once.
Veo 3.1 Fast (Google)Output secondsDrafting, social clips, high-volume video versioning.
Veo 3.1 (Google)Output secondsHero spots and finished broadcast-style deliverables.
Kling 3.0 (Kuaishou, via fal.ai)Output secondsCost-efficient social cutdowns; same product across many shots.
Kling 3.0 Pro (Kuaishou, via fal.ai)Output secondsA finished-looking single shot when Veo is too expensive and Standard is not holding up.
Runway Gen-4 (via fal.ai)Output secondsVFX-leaning brand films and shot-heavy storytelling where you expect to iterate on the motion.
H3 Max Reference (MiniMax, via fal.ai)Output secondsBlender-guided product films with recurring subjects. Output and reference inputs are billed separately; exact edit timing still needs review.
Seedance 2.5 (ByteDance, via fal.ai)Tokens · size + output + video inputA move between two frames you have already approved, and the cheapest Seedance route, since nothing is billed as input duration.
Seedance 2.5 Reference (ByteDance, via fal.ai)Tokens · size + output + video inputProduct-identity-critical ads: the packaging must stay exactly itself while the camera moves.
ElevenLabs Voice v3 (via fal.ai)CharactersAuditioning scene narration before picture, then placing the original voice files in the final edit.
ElevenLabs Sound Effects v2 (via fal.ai)Output secondsThe hero product sound: the crack, the pour, the seal breaking. The one noise the ad is actually selling, which a video model only ever approximates.
ElevenLabs Music (via fal.ai)Started minuteA separately directed instrumental music bed for an ad, with control over genre, instrumentation and musical arc.

Rates are configured estimates, not live quotes. Use the scenario controls above and verify the linked provider source before budgeting a production run.

My take on stills

Nano Banana Pro (Gemini 3 Pro Image)

The studio's default for detailed product stills and text-sensitive work. Review generated lettering against approved copy. Campaign Studio also offers locally typeset EN/FR layouts when exact words should remain editable.

Nano Banana 2 / Flash Image

The configured lower-cost image route for variations. Compare it against an approved hero and check the actual output; lower cost alone does not establish that it meets the brief.

Flux 2 Pro (Black Forest Labs)

A configured route for photographic hero imagery. Use the same product reference when comparing its material detail and lighting with another route; evaluate the result on the actual campaign rather than a general ranking.

Ideogram (not wired in, on the radar)

The typography specialist. If the studio's flyer/promo tile volume grows, a dedicated text-in-image model is the next evaluation. It is a good example of the continuous pilot-and-adopt loop.

My take on video

Veo 3.1 (Google)

The direct Gemini video route used by this studio, including Fast and Standard choices. Native audio is supported; reference-audio input and a native-audio off switch are not implemented on this route here. Inspect picture and sound together before choosing a take.

Kling 3.0 (Kuaishou)

Two configured fal routes for short product motion. This app offers fixed 5- or 10-second durations. Standard returns silent picture; Pro is treated as audio-on. Compare the exact scenario above rather than a blanket price multiple.

Seedance 2.5 (ByteDance) · reference-to-video

The configured multimodal route for the Blender-to-film workflow. Image, video and audio references each receive a role and positional token. The app supports up to 30-second output and limits reference output to 480p or 720p. A motion guide supplies direction; generated timing, product identity and text still need review. Video inputs affect the estimate as well as the output duration.

Runway Gen-4

The configured image-to-video route can be tested against the same product brief. Motion brushes and a general video-to-video editor are not wired into this studio; controls in a vendor's own application should not be confused with this integration.

Models outside the configured roster

A model appearing in a vendor announcement does not make it an available route here. Adoption requires an endpoint adapter, input validation, a cost estimate and a test against a real production brief. Keep projects exportable so the work survives a vendor change.

Ready / emerging / not yet viable

Ready to use

  • Stills generation & editing for adaptations and versioning
  • Text-in-image promo tiles (with a QA gate)
  • Short-form video (5–8s) with native audio
  • EN/FR headline localization with human review

Emerging · pilot now

  • Multi-shot narrative spots (Seedance-class models)
  • Automated brand-compliance checking
  • Video restyling of existing footage (WAN-class)

Not yet viable

  • Unreviewed publish-direct pipelines
  • Long-form (30s+) fully generated spots at broadcast QC
  • Photoreal likenesses of real people (policy, not tech)