Cloudflare's four transform modes vs Sume frames, detach and trim

Cloudflare Media Transformations has one endpoint with four output modes: video for an H.264 and AAC MP4, frame for a still, spritesheet for a JPEG with many frames, and audio for an AAC M4A. Sume splits the same work across three tools: video trim for a clip, video frames for stills, and audio detach for audio. Sume has no spritesheet mode, so a sheet means assembling frames yourself.

Cloudflare's list is from its Media Transformations page, read 2026-10-03. Sume's mapping uses video trim, video frames and audio detach.

Mode by mode

The names line up well enough to map one to one, but the output formats and limits differ.

What changes in practice

With Cloudflare you change mode and keep one URL pattern. With Sume you call a different endpoint per job, each with its own price and its own job to poll: trim $0.02, detach $0.01, and frames billed by Modal compute plus the platform fee. Each one needs the source to be on media.sume.com first.

Sume's frames return one durable image artifact per time, with a width and height for each, which is easier to place individually than a single sheet image.

Building a sheet from Sume frames

Ask for up to 24 frames with at[] or fps up to 2, then place them yourself in a layout you control. A timeline render can hold still images as static holds, and compose can place a still beside a video, but neither is a contact-sheet tool, so state your layout in your own code.

If you only need a poster, one at value is enough.

• format jpeg is the default, png is lossless.
• max_edge can clamp the long edge from 16 to 2160.
• Source must be 300 s or shorter.

Which route fits

If you already run on Cloudflare and need short previews, its single endpoint is convenient. If the clip is also going through filtering, detaching and timeline assembly, keeping every step on one hosted source is simpler, and Sume's separate tools each return a plain file you can pass to the next.

Cost and polling differences

Cloudflare's page states parameters, not a price per mode in the summary read here, so check current billing there. In Sume, each job has its own reserve and its own result: trim and detach are flat per job, while frames are billed by Modal compute plus the platform fee.

Plan one poll per job. Chaining three Sume tools means three jobs, each with an Idempotency-Key, which is worth it only when you need all three outputs.