HeyGen Pro 4K export at $49 vs Sume video upscale per clip

On HeyGen's pricing page the step from 1080p to 4K export is the step from Creator at $29 a month to Pro at $49 a month, so a 4K export gate costs $20 a month more. On Sume you can instead upscale a finished clip for $0.009 per input second, which is $0.09 for 10 seconds and $0.27 for the 30-second maximum. These are different products, so the numbers frame a decision rather than prove a winner.

What HeyGen's page says today

Credits are HeyGen's single currency for video and asset creation, per the page.

What Sume Video Upscale charges

A factor of 2 turns 1920 by 1080 into 3840 by 2160 by arithmetic. For a 10-second clip that costs $0.09, and the $20 gap between Creator and Pro would cover 222 such jobs.

• Public rate: $0.009 per input second.
• Up to 30 input seconds per job. Omit duration_seconds and the reserve assumes 5 seconds.
• Request fields: video_url (a public HTTPS URL), upscale_factor from 1.1 to 4, and enhancement_tier fast, standard or pro.

Why this is not a like-for-like swap

Upscaling enlarges an existing clip, so it cannot add detail the source never had, while an avatar platform that renders natively at 4K can. Sume's upscale also does not make or lip-sync an avatar. Use the comparison for one case only: you already have clips at 1080p or below and the only missing piece is a 4K delivery file. If you are choosing an avatar tool, judge it on the avatar, then add the upscale cost as a separate line.

Checklist before you upscale

Run one clip through before you queue a batch, and keep the original file. Upscaling is billed per input second, so a clip you trim to the useful 8 seconds first costs less than the full take. The job follows the usual async envelope, so submit with an Idempotency-Key, poll the job status, and fetch the result when it is ready.

• Confirm the clip is at most 30 seconds, or split it first with video trim.
• Give the endpoint a public HTTPS video_url; it validates that field as a public URL.
• Pick the factor from the target width, not from habit.
• Spot-check faces and text, which show artifacts first.