Palmier Pro, a video editing software for Mac, allows an external AI agent to control certain operations via the Model Context Protocol (MCP). A test conducted on two raw footage files shows mixed results: the agent proves efficient on repetitive tasks, but its editing choices remain less reliable.
- Palmier Pro permet à un agent IA externe de piloter certaines opérations de montage via MCP, mais il ne fonctionne que sur Mac Apple Silicon sous macOS 26.
- L’agent excelle sur les tâches répétitives : il réduit une interview de 18 min 43 s à 15 min 52 s en trois minutes.
- Ses choix de montage restent moins fiables, et une première instruction trop large a donné un résultat brut.
- Le doublage vocal est le plus convaincant, mais le flux de travail reste laborieux, et la transcription doit être vérifiée à l’écoute avant chaque coupe.
An installation reserved for recent Macs
Palmier Pro currently requires macOS 26 Tahoe or later, and a Mac with an Apple Silicon chip. Windows, Linux, and Intel Macs are not supported at this time. MCP is a protocol that allows an external agent to access a software’s commands: reading the project, cutting or moving clips, creating a timeline, launching an export. The connection is made in Palmier’s settings, and getting started is quick for those already familiar with this protocol, but less accessible for a beginner.
For the test, two files were used: 18 min 43 s of interviews and a second file of B-roll, i.e., illustrative shots. A first broad instruction, for a 90-second to 2-minute documentary on the making of The Mandalorian, produced a timeline of about 1 min 50 s. The result proved too raw: the talking heads were too long, the B-roll was sparse, and a video track covered the interview in places. A second, more precise instruction significantly improved the edit, after about 13 minutes of processing.
Palmier also allows cutting the video directly from the transcription. This feature is handy, but the English transcription sometimes lacks accuracy, with misrecognized words or fragments. It is therefore still necessary to listen to the audio before confirming a cut.
Music, dubbing, and vertical formats
The integrated chat generated a background music track announced in advance by Palmier, created with ElevenLabs Music v2.5. The track, about 96 seconds long, cost 20 credits and was placed on a separate track. The agent was able to verify its presence on the timeline but could not assess the quality of the mix by ear.
The vertical 9:16 version, lasting 52.4 seconds in 1080×1920 format, was deemed convincing for the selection of clips, subtitles, and use of B-roll. The cropping required two corrections: a first version removed the black bars but stretched the images, while a second attempt enabled proportional cropping without distortion.
For the dubbing, a 27.4-second segment with a single speaker was translated into French with ElevenLabs, for 23 credits. The most polished result of the test according to the report: the voice sounds natural and the lip sync works well on the clip. The workflow, more laborious than expected, constitutes the main limitation. The generative transition between two interviews, produced with Wan 3.0, yielded 2 seconds of video in 720×1280 for 20 credits, with interest deemed limited relative to the cost.
Cleaning up an 18 min 43 s interview was one of the most convincing tests. In about three minutes, the copy was reduced to 15 min 52 s, with 111 deletions: 20 long pauses, 80 hesitations, and 11 repetitions. The original remained intact on a separate timeline.
Offers at $29 and $69 during the test
Palmier Pro offers a free plan, without credits, including MCP and local transcription. The Pro plan is priced at $29 during the test, versus $49 outside the promotion, for 5,000 credits per month. The Max plan costs $69 during the test, versus $99 outside the promotion, for 12,000 monthly credits. According to Palmier, 5,000 credits correspond to approximately 333 images or 2 to 7 minutes of generated video, depending on the model, resolution, and duration. The three generations in the test consumed 63 credits, and cloud transcription also cost credits, with no specific amount stated by the publisher.
