Fred & Farid AI Agency Shift

The daily cost of manual handoffs

Fred & Farid teams once spent 14 days on a single 30-second brand spot. Storyboards moved between three external vendors, each charging per revision round. One missed sync between the motion poster and final cut added $4,800 in rush fees last quarter.

That exact workflow is the friction most mid-size agencies still face in 2026.

Where traditional pipelines break

Agencies keep separate logins for text-to-image, image-to-video, and lip-sync tools. Export formats rarely match. A 1080x1920 frame from one service lands at 4K in another, forcing re-renders that eat credits and calendar days. Fred & Farid tracked 22 hours of idle time per project just waiting on file transfers.

One dashboard with 50-plus models

They consolidated on Flixly. The same account runs Text to Video with Seedance 2.0, then refines frames in Image to Video using Kling 3.0 at 24 fps. Character faces stay locked across shots because Reference to Video pulls from the same seed image every time.

A single credit balance covers everything. No per-tool subscriptions.

Model mix that replaced three vendors

• Veo 3.1 for establishing shots (8-second clips at 1080p)
• Wan 2.7 for product close-ups with accurate lighting
• Gemini 3.1 Flash TTS for scratch voice before final Voice Cloning
• FLUX Kontext for style-consistent stills fed into Motion Poster

Average generation time per 30-second spot dropped from 11 days to 4.5 days.

Edge cases they still hit

Lip-sync accuracy falls when the source audio contains heavy accents. They route those files through Lip Sync Video twice with different mouth-shape weights. Complex camera moves longer than 12 seconds still need manual keyframe cleanup in external software.

Credit burn rises sharply on 4K exports; they keep 1080p internal reviews and only upscale final masters.

How the new workflow actually runs

1. Client brief lands in shared folder.
2. Quick thumbnail pass via Thumbnail Generator for approval.
3. Style frame locked with AI Image Generator.
4. Sequence built in Shorts Generator using First to Last Frame.
5. Final audio added with Text to Speech and cloned talent voice.

All steps stay inside one browser tab. The project folder exports a single timeline XML ready for the edit suite.

Limits that remain

Nano Banana Pro still produces occasional artifacts on reflective surfaces. They keep a 10% buffer in every budget for manual fixes. No model yet handles live-action crowd replication without extra reference footage.

Fastest way to test the same stack

Start a project in the Image to Video tool and run your first test clip with Veo 3.1 today.

Setting up reference consistency across multiple shots

Fred & Farid creatives start every sequence by locking a single hero frame generated in FLUX Kontext. That frame is uploaded once as the reference seed for every subsequent Reference to Video pass. They avoid re-uploading the same character from different angles; instead they generate three controlled variations (front, three-quarter, profile) in one batch and store them in a dedicated reference folder.

When a new shot requires a different camera height, the team crops the original seed to match the new framing before feeding it into the model. This single extra crop step prevents the face-locking drift that appears after the fourth or fifth shot in a 30-second spot. They also keep a running note inside the project of which seed ID produced the cleanest skin tones under the chosen lighting rig.

Export settings that minimize re-renders

Before any final delivery, the team runs a 15-second test clip at the exact resolution and frame rate the client will receive. They compare three output presets inside Flixly: 1080p 24 fps with light compression, 1080p 24 fps with high compression, and 4K 24 fps downscaled in post. The high-compression 1080p version consistently survives client revisions without introducing new artifacts, so they default to it for internal approvals.

A second table tracks which settings trigger extra credit use on longer camera moves:

They keep the final master at 4K only when the storyboard calls for heavy reframing in the edit suite.

Integrating exported timelines into Premiere or DaVinci

After the sequence finishes in Shorts Generator, the project folder produces an XML that already contains clip names, in/out points, and reference stills. Importing that XML into Premiere drops every clip onto the timeline with the original file names intact, so sound design can begin immediately. In DaVinci Resolve the same XML arrives with embedded LUT metadata from the style frame, saving the colorist the step of matching grades across shots.

One recurring friction is mismatched timecode when the scratch audio was generated at 48 kHz but the video at 24 fps. The fix is a single checkbox in the export panel that forces audio to 23.976 fps before the XML is written. After that checkbox was added to their checklist, the number of manual timecode corrections dropped from an average of nine per project to zero.

Daily checklist before handing off to sound

• Confirm every clip uses the same seed ID listed in the reference folder.
• Run a 1080p light-compression test render of the full 30-second spot.
• Check lip-sync on any line longer than four seconds by scrubbing in the Lip Sync Video preview.
• Export XML and drop it into the edit suite to verify clip order and duration.
• Archive the seed images and prompt text in the shared drive before closing the browser tab.

Following this list keeps the handoff to audio under 30 minutes even on days when three projects finish at once.

Reference folder organization for multi-project teams

Fred & Farid keeps one master reference drive per client. Inside each drive sits a folder named by project code, then subfolders for hero frames, angle variations, and lighting tests. Every seed image carries the exact prompt text in its filename so anyone can recreate the starting point without opening the original generation. When a new brief arrives, the team first checks whether an existing hero frame from a prior campaign can serve as the base; they duplicate it into the new folder rather than regenerating from scratch.

They also maintain a simple CSV log that records seed ID, model used, lighting rig, and any noted artifacts. This log travels with the project XML so editors and sound designers know which frames were locked early. A second column tracks which seeds survived client revisions, helping the group retire frames that consistently required extra cleanup passes.

Audio-video sync adjustments in export settings

When scratch voice lines come from Voice Cloning at 48 kHz but the video timeline runs at 23.976 fps, the mismatch shows up as drifting lips after the third or fourth line. The fix is a single dropdown in the export panel labeled "Force audio to timeline rate." Selecting it writes a new WAV at the exact frame rate before the XML is generated.

They test the setting on every project that mixes cloned dialogue with music beds. One recent 45-second brand film needed three separate voice takes; applying the rate lock once eliminated nine manual nudge corrections in DaVinci. The team now adds this step to the daily checklist whenever the project folder contains more than two cloned lines.

Scaling the workflow for campaigns with 10-plus deliverables

When a single campaign requires 12 deliverables instead of one 30-second spot, the team batches the style-frame step first. They generate 18 hero frames in one Text to Image session, then assign three frames per deliverable based on aspect ratio needs. Each deliverable gets its own subfolder inside the Shorts Generator project so clip names remain unique when the final XML is imported.

Credit tracking moves to a shared spreadsheet that lists every deliverable, its target length, chosen model mix, and running total. After the first four deliverables they pause to review which preset consumed the fewest credits per finished second; that preset becomes the default for the remaining eight. The same sheet records which seeds produced the cleanest skin tones so later deliverables reuse them without new tests.

They close each campaign by moving the final seed folder into an archive drive and updating the CSV log with total credits and any new artifacts observed. This archive becomes the first place the group looks when a follow-up brief arrives six months later.