fal Agent campaign workflow: references, revisions and current cost scenarios

Full evidence

fal Agent makes this workflow easier to follow: establish one strong reference, approve a production plan, build the campaign frames, and animate the selected images. The winter-coat demonstration shows useful human control at each stage. One frame needs a replacement, and the first video clips need a duration revision. Keep those review steps in your own plan.

This companion follows Kingy AI’s October 2 video, its transcript, and targeted frame checks. The recording shows a paid-promotion label. The sponsorship terms, any affiliate commission and any sponsor review of the video are unverified. The creator’s existing fal link remains available on YouTube. No new generation test was performed for this article.

Watch the useful steps

• 1:49: Sign in and open fal Agent
• 2:03: Give the winter-fashion brief
• 2:36: Review the proposed models and approve the plan
• 2:57: Lock the hero reference
• 3:10: Replace a weak campaign frame
• 3:23: Choose the frames to animate
• 3:53: Request two ten-second clips
• 4:23: Ask for the generation-cost breakdown
• 4:44: Start the coffee explainer
• 5:37: Add music and sound effects

Follow the campaign from reference to revisions

Start with the coat, the person wearing it, the environment, and the intended visual style. In the recording, the creator asks for a premium winter campaign with a long, elegant coat and snowy settings. fal Agent proposes the production sequence and selects Nano Banana Pro for the hero image. The creator inspects that image before approving it as the reference.

The next step produces four campaign frames. The fourth needs a different prompt. That correction matters: accepting a plan does not mean accepting every generated asset. Inspect the coat’s shape, the subject’s identity, the hands, and the scene before spending on animation. These are suggested review criteria, rather than separately measured scores from this demonstration.

Choose the frames with a clear motion idea. Here, the agent suggests a street walk and a close collar portrait. The creator accepts those choices, then asks for ten-second versions of both clips. The inspected recording identifies the video model as H3 Max Image to Video and shows a 1080p revision. It does not establish that other video models were tested.

Keep the reference and the approved directions attached to revisions. Ask what changed, what stayed fixed, and which new generations incur charges. A replacement clip can add to your spend while the earlier attempt remains part of the project’s cost.

Prompts you can adapt

The following briefs summarize the requests in the video. They are editable starting points, not exact hidden model prompts or a promise of matching output.

Create a premium winter fashion campaign around a model wearing a long, elegant coat. Use snowy settings and a consistent high-end fashion style. Establish the coat and model first. Propose a production plan for several campaign images, then suggest which approved frames to turn into video. Ask me to approve each stage.

Replace the fourth campaign frame with a different composition. Keep the approved coat, subject and visual direction. Explain the proposed change before generating it.

Make both selected clips ten seconds long, using the same source frames and approved prompts. Show the expected extra generation cost before running the revision.

For the second project, the creator requests a fifteen-second coffee explainer covering growing, roasting, and brewing. The recording shows the illustration sequence, a resulting animation, and an audio revision. The transcript does not name the animation or audio models, so those identities remain unverified.

Create a fifteen-second coffee explainer covering grown, roasted and brewed. Use clean infographic illustrations, warm coffee colours, readable typography and smooth transitions. Show the stages for review. After the visuals are approved, propose subtle music and sound effects.

Tools, models and the sign-in provider

Why the recorded $3.30 estimate differs from published rates

The cost frame totals six images at $0.15 each and thirty generated video seconds at $0.08 per second. That gives $0.90 + $2.40 = $3.30. The thirty seconds include the earlier two five-second clips and the later two ten-second revisions. Counting only the final twenty seconds would omit the first attempts.

The same frame says its $0.08 rate was a launch promotion ending September 30. The video was published on October 2; its filming date is unknown. The stated promotion had therefore ended before publication. The frame describes these numbers as per-run estimates, not settled charges. Treat $3.30 as the recorded agent estimate. This mismatch does not establish what the account was actually charged.

When checked at 05:43 UTC on October 3—October 2 Pacific time, after the video was published— fal’s exact H3 Max image-to-video page listed 1080p at $0.096 per generated second, under a different promotion ending October 15. It listed $0.16 per second after that promotion. Nano Banana Pro’s generation page listed $0.15 per image, double for 4K, and an extra $0.015 when web search is used.

Using those published endpoint rates, six standard images and thirty video seconds produce a $3.78 planning scenario. Using the listed later video rate produces $5.70. These calculations exclude any Agent-specific billing differences, search add-ons, audio, tax, credits, or plan discounts. Confirm your project’s approved estimate and account billing before running it.

The video description refers to side-by-side model comparison. The inspected cost frame explicitly says this campaign used one chosen model at each stage and ran no model comparisons. This page therefore makes no A/B quality, speed, or cost claim.

Plan attempts, not only finished assets

Change the counts in the calculator to budget new images, repeated takes, and total generated video duration. The calculator uses the reviewed published endpoint rates while they remain current. Enter the image count, total generated video seconds, a revision reserve and any 4K image requirement. If its pricing evidence expires, it withholds the estimate until the rates have been reviewed again. Compare its estimate with the recorded $3.30 breakdown above; they use different video rates.

Standing Answer: should you animate every campaign frame?

Approve your reference and review the stills first. Animate the frames with a useful motion plan, then budget revisions as additional attempts. This demonstrated workflow supports that sequence; it does not establish a universal cheapest model. Read the maintained answer on planning an AI campaign and its generation budget for the current source checks and calculator assumptions.

What this demonstration leaves open

The retained evidence covers two creative projects and targeted frames. It does not establish repeatability across a larger sample, failure rates, output rights for a particular client project, settled charges, or comparative performance. Before client delivery, review the finished files, reference permissions, branding and the provider’s applicable terms.

Official references: fal Agent, Nano Banana Pro editing, H3 Max Image to Video, and GitHub’s product explanation. The creator’s video remains the source for the recorded workflow and historical estimate.