Subnet 99: Leoma
Leoma is Bittensor Subnet 99. Public Leoma materials describe it as an AI video subnet where miners produce short videos from image-and-text challenges and validators rank the outputs for on-chain weights.
What Leoma Provides
The Leoma documentation frames the subnet around Text-Image to Video work. Validators sample a first frame and a prompt from source clips, miners return generated video clips, and the resulting outputs are compared against the challenge rather than against a generic video-quality score alone.
That task shape matters because the first frame anchors the generated clip. A strong miner output has to preserve the input frame, follow the prompt, keep motion coherent over time, and avoid visual artifacts. The Leoma README describes this as a multi-aspect evaluation with pass or fail results, followed by winner-take-all ranking for the round.
Leoma therefore evaluates conditional video generation rather than open-ended media creation. The subnet’s current scope is TI2V, while the public docs describe text-only video and image-only video modes as planned extensions rather than the current production task.
Miner and Validator Roles
The README describes miners as making a TI2V model available for challenge evaluation. The miner’s useful contribution is the model’s ability to turn a frame and prompt into a short video that survives the subnet’s evaluation criteria. The mechanism rewards outputs that win the current comparison, not simply the act of appearing on the subnet.
Validators supply the challenge context, evaluate the returned clips, and use the current ranking to set subnet weights. The public docs describe ranking as pass-based and winner-take-all, so a round is about selecting the strongest valid video output for the sampled task rather than distributing weight across many partially successful miners.
Evaluation Context
Leoma’s evaluation context is deliberately narrow. The docs say the active task type is Text-Image to Video, so the subnet’s current production task is conditional video generation from both image and text inputs. The same challenge can test several qualities at once: whether the first frame remains recognizable, whether the prompt is followed, whether motion is temporally plausible, and whether obvious visual defects appear.
The README also describes a model reference and serving target being recorded on-chain. That detail matters because validators need a stable way to know which miner model they are challenging, while the scoring process still depends on the generated clips themselves. The model reference identifies the participant’s candidate system; evaluation decides whether that system produced the best video for the round.
On-Chain Identity
Live SN99 data is available on TaoStats. The source-backed video generation and evaluation details in this article come from the public Leoma repository and documentation rather than from live identity fields.
Relationship to Yuma Consensus
Subnet 99 uses Yuma Consensus to convert the video-generation ranking weight vectors that validators submit into the emission shares distributed to miners and validators within the subnet each tempo. The Yuma Consensus documentation describes how validator weight submissions are aggregated into consensus weights for each miner registered on the subnet.
In Leoma’s context, validators sample a first frame and text prompt from source clips, evaluate miner-generated video outputs on frame preservation, prompt adherence, temporal coherence, and visual quality, then set winner-take-all weights for the round based on which submission passes evaluation and ranks highest. The Emission documentation describes how those consensus weights determine each participant’s share of the subnet’s accumulated emission each tempo.
Development Stage Context
The Introduction to Bittensor describes subnet development as moving from localnet to testnet and then mainnet. For Leoma (SN99), that sequence changes how readers should interpret AI video generation examples and winner-take-all scoring outcomes.
In localnet, Leoma-compatible miners and validators can be developed and tested in an isolated environment. Localnet video evaluation results and emission outcomes do not represent production subnet performance.
On testnet, Leoma-compatible text-image-to-video model submissions can be exercised in a shared, non-production network. Testnet video ranking results and validator weights are separate from mainnet subnet state.
On mainnet, Leoma (SN99) is the live production subnet where miners generate short video clips from image-and-text challenges and validators rank those outputs to determine real Bittensor emissions. The Leoma repository describes the mechanism that applies on the production network.
The Bittensor Networks reference separates mainnet, testnet, and localnet. A video generation result or emission outcome from one environment should not be read as representing production subnet performance in another environment.
Netuid 99 Identifies the Subnet On-Chain
Bittensor assigns every subnet a unique numeric identifier called a netuid, and Subnet 99 is the subnet registered at netuid 99 (Glossary: Netuid). The Understanding Subnets reference explains that each subnet runs its own incentive mechanism while sharing the same underlying Subtensor chain, so the netuid is the stable handle that distinguishes Leoma from every other subnet.
For a reader, this means “Subnet 99” and “netuid 99” refer to the same on-chain slot. A claim about Leoma should be tied to that netuid rather than to the registered name alone, because the name field can be changed on-chain while the netuid stays fixed.
Reader Boundary
Subnet 99 Leoma should not be read as generic Bittensor subnet documentation, a consumer video-generation product, or proof that any generated clip is original or rights-cleared. It names one subnet’s AI-video competition where miners run text-image-to-video models and validators score the generated clips with winner-take-all weights each round on netuid 99 (Understanding Subnets, Glossary: Netuid).