My latest short film is heading for a million views on YouTube in less than 24 hours.
I know many filmmakers working with AI aspire to that. But we need to separate real results from misleading promotion, praise from bots, and applause from the same familiar circle.
In my experience, the numbers that make this work sustainable are on YouTube. On X especially, I see a lot of appearances and a small circle congratulating one another, sharing each other's work and making noise. Building an audience that can sustain your films is a different challenge.
Instagram, TikTok, Facebook and X aren't where I find that profitability. YouTube is.
Before you have sponsors behind you, you should be able to turn a profit without them. If you depend on a brand to keep creating, that brand has too much power over what you make. I don't want to end up selling my soul for pocket change.
Over the last 28 days, my YouTube channels have generated almost 200 million views. I make films for that audience, and I try to give them new stories and new characters.
If we use this technology to repeat what Hollywood was already doing, what are we adding? Just changing who produces it?
For me, this is about innovating, developing my own style, refining it and continuing to take risks. Trying things, making mistakes and learning.
Repeating myself might be comfortable, but it would cost me in the long run. If I stop learning, I stop improving. And at the speed this field is moving, standing still means falling behind.
Here's a 4-minute, 34-second scene from Lion Island where I tried to maintain the strongest possible continuity between shots.
If you spot mistakes, tell me where and why. I want to keep improving. Thank you.
HOW MUCH DOES IT ACTUALLY COST TO MAKE JUNKYARD KING? 💸
Everyone seems to be asking the same question lately: how much does AI filmmaking actually cost?
I’ve seen quite a few breakdowns circulating, with wildly different numbers, so I thought I’d share ours at @thisisoddkid with as much transparency as possible.
We’ve been producing Junkyard King as a serialized animated project, with episodes ranging from around 10 to 16 minutes, and a team of just two people.
Here’s what we’re actually spending.
1. COMPUTE: $1,000–$1,500 PER FINISHED MINUTE
This includes video generation and image models, although image generation is relatively inexpensive compared to video.
We try to be smart about it. Lots of draft generations until we find the right recipe, then we spend more on the final shots.
For a 16-minute episode, that means roughly $16,000–$24,000 in compute alone.
2. SOFTWARE & SUBSCRIPTIONS
Here’s our current production stack, with approximate US pricing:
Adobe Creative Cloud: ~$70/month for Premiere, After Effects and Photoshop.
Claude Max: $200/month, mostly for MCP workflows and production tooling.
ChatGPT: ~$20/month. We still use it occasionally. Different models, different flavors.
ElevenLabs Pro: ~$99/month for voice generation.
Suno Pro: ~$10/month for music.
Figma / FigJam: ~$5–$20/month per paid seat, depending on the plan. Our shared production boards.
Google Drive / Dropbox: ~$10–$20/month per service for storage and exchanging files.
Topaz Video AI: Paid license, with pricing depending on version and upgrade terms. I’m still using Artemis on an older version.
Stock sound effects: Mostly free, with occasional paid assets.
Additional After Effects plugins: A few one-time purchases.
That’s roughly $415–$460/month in recurring software subscriptions, depending on storage and Figma plans, excluding Topaz, plugins and occasional assets.
Not exactly the biggest part of the budget.
3. LABOR: THE BIG UNKNOWN
We’re currently making Junkyard King with two people. I’m working 3/4 time on it, and my collaborator is half-time.
Some episodes I handle lots of the shots, others she takes on more. It really depends on the scenes we’re each tackling and how they turn out.
And this is where I think these breakdowns become a little misleading.
How much do you pay yourself? How much does your team charge? How much time are you willing to spend fixing something that technically works but isn’t quite right?
Those decisions can completely change the economics.
4. THE PART NOBODY CAN REALLY PREDICT
Our ratio of selected shots to generated shots varies enormously.
Some scenes are relatively straightforward. Maybe 5 generations for 1 usable shot.
Others are an absolute nightmare, closer to 30:1 before we figure out what works.
It’s a LOT of trial and error.
For example, the episode we’re currently developing takes place largely inside a shop, with four characters sharing a relatively small space.
Every movement of one character needs to be reflected in the rest of the scene. Positions, eyelines, interactions, continuity. And you obviously can’t have four characters standing in the same spots talking for ten minutes.
Suddenly, a scene that sounds much simpler than a massive action sequence becomes a production headache.
And that’s the thing.
A minute of animation is not a unit of complexity.
One minute can come together surprisingly quickly. Another can eat days of your life because two characters refuse to interact properly.
This is why I struggle with the idea that we can establish a universal cost-per-minute formula for AI filmmaking.
The tools have made production dramatically more accessible. I wouldn’t be making Junkyard King at this scale without them.
But the real cost depends enormously on what you’re trying to achieve, the quality you’re willing to accept, and how much time you’re prepared to spend getting there.
Compute is relatively easy to calculate. The cost of caring is harder. 😅
EveryFrame AI Film Contest No. 01 is open 🎬
Wildlife Documentary Competition: a 2-min narrated wildlife film, every shot on EveryFrame.
🏆 $5,000 Prize value to one winner
🗳 Community votes
📅 Closes 27 Nov, 12:00 UTC
Make it: https://t.co/1KIkYpgWUL
Enter 👇
https://t.co/62FiudovuD
We’ve made a community film on Bittensor's history, from the very first block to the first halving and the network today.
The Open Network: From Kusanagi to Finney
Every shot made in EveryFrame.
The dates and figures are real; the faces and dialogue are imagined.
#Bittensor #TAO #SN117
most subnets run in the same issue after starting to print off-chain revenue:
what's the incentive to share revenue with alpha token holders?
since dTao leak a forced incentive,
only the market create a incentive.
bittensor:native
Bittensor 24H Market Snapshot
🔹 RedTeam (SN61): Ethical hackers compete to break and harden bot/fraud detection systems, feeding real-time improvements into commercial cybersecurity products.
🔹 Everyframe(SN117): One studio for AI video, image, and audio across dozens of models, with miners competing to drive prices down.
🔹Hippius (SN75): Blockchain-backed decentralized cloud for storage, VMs, and CDN with confidential computing and transparent billing.
🔹TPN (SN65): Competitive AI model compression engine that rewards miners for shrinking models while preserving intelligence, verified on-chain.
🔹 Lium (SN51): Peer-to-peer GPU rental platform (Airbnb for GPUs) with hardware verification, high utilization, and revenue sharing for providers.
🔹engy (SN53): Lowest-cost verified inference for frontier open models (e.g. Kimi K3) running on consumer GPUs with cryptographic proofs.
🔹 Targon (SN4): Confidential AI compute marketplace offering VMs, bare-metal servers, and sandboxes, funded in part by organic revenue buybacks.
🔹SayGM (SN28): Private TEE-routed gateway to Claude, GPT, Gemini, and open models—one API key, competitive pricing, prompts stay private.
🔹Minos (SN107): Merit-based genomic variant-calling competition where miners maximize accuracy on sequencing data for rewards.
Source: SubnetRadar
GPT Image 2.6 is now on EveryFrame, and it makes GIFs.
One prompt, 12 frames of motion.
• Animated GIFs
• Photo to GIF
• Circle-to-edit: circle it, say it, everything else stays put
• Exact text, even when the letters move
Stills from $0.04.
https://t.co/XrXRzwBbM0
#Bittensor #contentcreators #AIFilmmaking #SN117
I found the LOWEST cost AI API provider
It’s crazy, this AI API can literally cut any company’s bill by up to 66%
what i’m paying:
seedance 2.5 → $0.07/s
kling 4.0 → $0.07/s
wan 3.0 → $0.015/s
& more
how much cheaper than:
kie → 50%
higgsfield → 66%
fal → 67%
same models. same quality. just fair prices.
RT & drop “API” below and i’ll send the link
What if intelligence had no owner?
From one idea in 2019 to 128 subnets: the Bittensor story so far.
Kusanagi. The halt. The restart. Finney. Upgrades without stopping. The first halving.
All footage and the score made on @EveryFrame, Bittensor subnet 117.
#bittensor $TAO
SN117's unused emission now funds AI filmmakers instead of being recycled.
Miners are still paid first. Here's how it works 👇
🔹 What doesn't change
Every epoch, miners are paid before anything else: serving fees plus a 10–50% quality bonus based on the quality of their output. The treasury only receives what's left after that. It can never come out of miner rewards.
🔹 What changes
Until now, emission not earned by verified work was recycled. From now on it goes to a single EveryFrame Treasury Wallet (UIDs 156 · 157).
🔹 What the treasury is for
One thing: the EveryFrame AI Film Contest. Creators make films on EveryFrame, the community votes for the winners, and the treasury pays the prizes.
🔹 Why a treasury instead of recycling
Recycling has its place, because it keeps unused emission out of circulation. We think SN117 gains more by putting that emission to work. A film contest brings creators to EveryFrame. More creators mean more real usage, which means more demand on the network and more serving fees for miners. It's a reinvestment in growth.
🔹 How to verify it
✅ One known wallet, the same UID picked by every validator
✅ Winners decided by community vote, not by the team
✅ Every award approved and paid on chain
✅ The treasury dashboard shows the split, the balance and every payout, every epoch
Every number can be checked on chain.
Track every epoch here 👇
https://t.co/meQFNFtfrD
Questions or pushback? Join our Discord channel and talk to the team.
Treasury wallet address
https://t.co/zb7Q02POVc
#Bittensor #AIFilmmaking #ContentCreators #SN117
Get more for less. This time, less = $0
Earn free EveryFrame credits:
Quote us → 1,000 credits
Reply to us → 500 credits
Invite a friend → 1,000 credits
Copy your code, post, paste the link. Credits land instantly.
👉 https://t.co/NY8xyCSm2l
#Bittensor#Aifilmmaking #ContentCreators #SN117
Halloween on Mars looks clean.
And yes - EveryFrame has confirmed they’ll be using real subscription revenue for alpha buybacks once it starts flowing.
Still early, but they’re already planning to put money back into the token.
#Bittensor#TAO#SN117#AI#AIvideo#crypto $TAO $SOL $COIN $HOOD #Solana #Coinbase #Robinhood #Robinhoodchain
https://t.co/JiXZjiVClR
EveryFrame (#SN117) is quietly assembling the full AI film stack and $TAO is the engine underneath it.
AI filmmaking today is fragmented: video here, images there, voice somewhere else, editing in another tab. EveryFrame is collapsing that mess into one script to screen studio: generation, direction, editing, sound and export under a single workspace and credit balance.
One studio, every format
@xeveryframe brings leading models: MiniMax, Seedance, Google Veo, Kling, image tools and ElevenLabs audio into one place, with visible generation costs and no need for separate provider accounts.
It goes beyond clips: Director tools handle background swaps, people replacement, lighting, lip sync, sound, and upscaling, while Cinema Studio supports screenplays, consistent characters and locations, storyboards, shot planning, timelines and final export.
That is the difference between an AI toy and an AI production suite.
SN117: the competitive serving layer
The real asymmetric bet is Subnet 117. Independent miners compete to serve eligible AI generation jobs, using existing provider accounts and lightweight TEE secured workers with no GPU fleets required.
Every job produces a signed receipt linking the request, model, and output, giving EveryFrame verifiable execution and validators a basis for rewarding reliable miners.
If more Studio workloads route through SN117, miner competition could compress serving costs and eventually make frontier AI creation cheaper for creators and developers.
Why the bull case works
Real product, real demand: EveryFrame is not a mining only subnet; it has a creator facing studio, subscriptions, credits and a developer API.
Capital light network: Miners contribute access and reliability, not expensive hardware, broadening participation.
Verifiable by design: Signed receipts and trusted execution make the marketplace accountable rather than opaque.
Multiple revenue paths: Creator plans, credit top ups, $TAO/crypto payments and pay per render API usage create several ways to monetize demand.
Early but live: #SN117 began serving verified mainnet jobs on September 28, 2026, with public miner onboarding already open.
The thesis
EveryFrame’s opportunity is to own the creative front end while #SN117 becomes the open, competitive infrastructure layer beneath it. If miner discounts increasingly reach Studio credits, EveryFrame could offer a rare combination: premium AI models, complete production tooling, transparent pricing and verifiable execution.
The execution risks are real: model coverage, reliability and passing savings to customers still need to be proven.
But the setup is clean: one studio for creators, one market for miners, one receipt for proof. EveryFrame is not just another AI video platform; it is building the OS for AI native filmmaking on Bittensor.
This was a video generated by a community member going ahead and testing @xeveryframe studio.
A comparison being made is against Higgsfield AI - the creative suite allowing for image and video generation.
Everyframe Studio is cheaper. The video cost > $1.
One thing that isn’t scary on Halloween 🎃
EveryFrames Prices!!!
Pay less. Make more. Only on EveryFrame !!
https://t.co/cAtNHwhnS4
#Bittensor#ContentCreator#AIContent#SN117