Feat/veo per tier pricing updates - #420
Merged
Merged
Conversation
Veo manifests never expose the model variant, so any variant-based floor is an assumption, not a detection. Drop the guessing: Veo videos are priced by delivered tier alone (_VEO_TIER_PRICES, audio-independent) — 480p $0.07 and 720p $0.152 match Seedance fast (same resolution, same multiplier), and 1080p prices at $0.40/s, the veo-3.1 standard 1080p rate, reflecting that Veo's real price rises with resolution. Multipliers: 1.53x / 2.25x / 3.65x. MODEL_TIER_PRICES stays as the live Veo price reference and becomes the pricing source again if variants ever become distinguishable.
… freshness Three anti-gaming changes to generative rewards, plus a scoring rebalance: - Classify video resolution tiers by total pixel count instead of minimum dimension. Min-dim classification priced square 640x640 output (480p pixel count and provider cost) as 720p; the top two miner groups submitted 1:1 aspect exclusively to collect the 3.65x multiplier at 480p cost. Thresholds sit at ~85% of nominal 16:9 pixel counts. Image tiers keep min-dim classification, which is their definition. - Discount video rewards 0.6x per tier delivered below the requested tier (VIDEO_UNDERSHOOT_TIER_DISCOUNT). min(observed, requested) pricing alone made the cheapest tier the best reward-per-dollar under the sqrt taper, so 99% of top-group videos ignored 720p/1080p requests. Undershooting still pays (degrade-not-fail preserved) but honoring the request is now always the highest-reward choice per challenge slot: on a 1080p request, 1080p Seedance earns 4.0x, 720p 2.19x, 480p 0.66x. - Reject miner media whose C2PA manifest was signed before its challenge was issued (10 min clock-skew allowance). Signing time comes from signature_info.time, falling back to the signed c2pa.created action's `when` (ByteDance). Closes replay of hoarded or scraped C2PA-signed content that outlives the dedup window. Verified against 60 recent live submissions across all three active providers: all carry timestamps, all signed 13s-15min after issuance, zero false rejections. - Rebalance generator scoring weights from 0.12/0.88 to 0.30/0.70 image/video so image challenges are no longer economically ignorable (miners currently decline them wholesale).
dylanuys
approved these changes
Jul 23, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
High Risk
Changes which miner submissions pass validation (C2PA timestamps) and how video/Veo rewards are computed, directly affecting validator weights and miner incentives across the subnet.
Overview
Release 4.9.5 tightens generative validation and reward economics, and shifts default modality scoring to 30% image / 70% video.
C2PA freshness: Manifest signing time is parsed from
signature_info.timeorc2pa.actionsand exposed on verification results. Accepted media must be signed after the challenge was issued (with a 10-minute clock-skew allowance), blocking replay of older signed content.Video resolution: Tier classification switches from minimum dimension to total pixel count, closing square-aspect mis-tiering.
tier_shortfalldrives a 0.6× per-tier multiplier discount on video rewards when delivery is below the requested tier (on top of existing min(observed, requested) pricing).Veo (C2PA-blind) pricing: Blind Veo generations are priced by delivered tier via
_VEO_TIER_PRICES(audio-independent), replacing the prior single floor-model +MODEL_TIER_PRICESlookup path.Reviewed by Cursor Bugbot for commit 922689a. Bugbot is set up for automated code reviews on this repo. Configure here.