Decision session · 20 July 2026
Video creative offer: Luma replaces Pika/Fal
Settled call on how Piexels sells, prices, and produces short-form video. One generation endpoint, one repeatable workflow, and three client-ready packages.
Luma Ray3.2 becomes the canonical video engine for every Piexels video deliverable. Pika and fal.ai are retired from the standard video offer. Runway stays documented as the escape hatch; Higgsfield leaves the production path.
New standard
Luma Agents API
Ray3.2 for video. UNI-1 for keyframes and stills. One REST surface, keyframe control, product swap, motion transfer, and a shared agent board.
Pika/Fal no longer standard.
What stays
Kimi + Remotion
Kimi turns the brief into a Luma prompt pack and orchestrates the job. Remotion adds the brand text and motion graphics spine.
Runway remains the documented runner-up.
Cost basis: $0.30 per 5-second 720p Ray3.2 clip and $0.0909 per 2K UNI-1 image. A typical 20-second property reel costs roughly $1.20–$2.00 in media before editing, QA, and delivery.
Context
Three things forced this decision now.
- The platform endpoint decision is already made. On 16 July the hub picked Luma Agents API as the canonical generation endpoint for images and video. Keeping Pika/Fal as the video bridge contradicts that call and adds a second credential, second budget cap, and second failure surface.
- The ProPick walkthrough proved the bridge is workable but not productizable. The fal.ai Seedance 2.0 reference-to-video test produced a usable 10-second clip, but it required a topped-up fal balance, manual storage uploads, and one-off metering. That is a prototype lane, not a sellable workflow.
- The marketing suite needs one video model clients can buy. The previous design mixed Higgsfield Ultra, fal.ai Kling bridges, Pika atomics, and Remotion. Each layer has its own cost, quality variance, and operator. Clients will not pay for fragmentation.
The settled call
Luma replaces Pika/Fal for video
Use Ray3.2 for all standard video generation: property reels, product motion ads, social clips, and walkthroughs. Use UNI-1 for keyframes, stills, and reference images. Upload campaign assets once and chain from Luma file IDs.
Pika/Fal leave the standard path
Pika remains useful as an effects specialist for one-off experiments only. fal.ai becomes a fallback bridge for legacy tests, not a production dependency. Neither appears in client proposals or standard operating docs.
The sellable unit is not the model. The sellable unit is a finished reel, with a fixed scope, a fixed price, and a brand-safe QA gate.
Sellable offer
Opening prices. Cost basis is public Luma media pricing; margin covers orchestration, QA, Remotion finishing, one revision round, and delivery in 9:16 and 1:1 MP4. Adjust after the first three closed sales.
- Up to 6 keyframes or reference images
- Cinematic camera path from Luma Ray3.2
- Remotion brand text + end frame
- Delivered 9:16 + 1:1 MP4 + thumbnail
- Best for property listings or product drops
- Shared brand pack loaded once
- Priority queue + consistent look
- Delivered batch every Friday
- For active campaigns and growth clients
- Camera + motion style locked per brand
- Weekly or on-demand batch delivery
- Direct-to-Telegram approval loop
Add-ons: product swap or motion-transfer variant (+€35), 16:9 cut (+€15), licensed music or voiceover (client cost + €25 handling), extra revision round (+€45).
Workflow
One repeatable path from brief to delivered video.
Kimi reads the active brand pack and campaign brief, then outputs a Luma prompt pack: keyframes, camera direction, motion transfer notes, and aspect ratios.
Luma Agents API runs Ray3.2 async jobs. Start/end frames and up to 16 keyframes keep the camera and subject coherent across clips.
Every clip passes a machine vision check: no morphing, no wobble, structure preserved, brand-safe colors, no AI tells. Failing clips regenerate automatically.
Brand typography, end frames, and captions are added in code. The same template reproduces the look across every reel for the client.
Deliver to the review hub. Client leaves per-asset feedback or approves. Feedback updates the brand pack for the next run.
Final MP4s go to the client, to Metricool, or directly to ad accounts. Every output is tagged with model, prompt hash, and source file IDs.
Luma vs. Pika/Fal: the call
| Dimension | Pika/Fal path | Luma path | Decision driver |
|---|---|---|---|
| API surface | Multiple bridges, accounts, and keys. | One Agents API for images, video, edits, and reframe. | Less operational surface. |
| Motion control | Limited keyframes; strong effects only on Pika. | Up to 16 keyframes, camera/motion transfer, product swap. | Property and product coherence. |
| Predictable cost | Fal metered tiers and balance top-ups vary. | $0.30 per 5s clip, $0.0909 per 2K image. | Fixed packages can be priced. |
| Workspace | No shared agent board across image and video. | Agent-native board for shared references and revisions. | Client and team collaboration. |
| Brand memory | Each job starts from scratch. | Files API stores assets and chains from IDs. | Compounding quality and speed. |
| Best use | Pika = effects; Fal = bridge experiments. | Canonical production video and images. | One standard, one escape hatch. |
Recommendation and next move
Standardize the video offer on Luma Agents API now. Remove Pika/Fal from client-facing workflows. Use the single-reel and monthly-pack prices as the opening offer. Keep Runway documented for the rare case where breadth beats control.
- Produce the first Luma reel for ProPick within 48 hours. Use the already-enhanced Whizdom Essence photos as keyframes to prove Ray3.2 can match or beat the fal Seedance walkthrough on quality and cost.
- Update the marketing suite operating doc to reflect Luma as the production video model and retire the Higgsfield/Pika/Fal fragments.
- Open the first paid pilot with one existing client at the €95 single-reel price, then close three pilots before adjusting the pack pricing.