Mark Ledford (@fishmongr) header
Mark Ledford (@fishmongr) profile picture

Professional engineer, creative technologist, and sometimes Twitter lurker. CTO, Team Sogni AI ✨

Seattle
Joined July 2008
Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)Photo by Mark Ledford (@fishmongr)
You ever have a cool idea for a movie that you'd love to see get made? We are already in an era where you no longer have to wait, you can make it yourself. Check out this trailer for Sogni Meme World, a fun little movie I made with my AI Agent through a couple conversations. It mixes as many memes and meme culture references into one experience as I could fit while telling a legitimate hero's journey based loosely on the plot of the 2006 Action/Thriller movie "Crank" with Jason Statham. If you like the trailer, try experiencing the game at sognimemeworld.com
1
1
4
527
The crazy part: The Agent not only built the trailer, it built the game. It registered the domain and deployed it too. Mostly in two evenings. It's powered by Sogni Worlds Kit, an AI Agent Skill you can load into any agent like Claude Code or ChatGPT Codex, or Hermes etc. Once the Agent Skill is installed you just ask your agent to build what you want. I've been building a lot of really cool apps 2+ years into our Creative AI startup Sogni AI and our technology has only continued to accelerate at a pretty phenomenal rate. What wasn't possible last year, or last month or last week, is possible this week. The era where you had to learn tedious interfaces or even learn how to write good prompts or prepare a boatload of assets can be over if you want. You can bring assets or it can generate them all using 200+ open-weight models. If you want to get fancier, like I did here, you can have it use a frontier video model like Seedance 2.5, uncensored on Sogni. You just have to power the Agent Skill with a good LLM model. My favorite by far: Claude Code Opus 5.5 which I used here with our skill. We've created a bunch more practical demos of what you can do with our interactive video storytelling skill here worlds.sogni.ai
97
Mac Studio M5 Ultra. 512gb. Imagine what you could build with that AI power. But... what if you could blow yourself away without waiting until December delivery? And a $20k hardware bill. What if you could get blown right now. For $20. I went to sleep last night with an idea for a crazy new 4chan / Reddit inspired pitstop in my Sogni Meme World interactive video game. I talked to my @Sogni_Protocol Creative Agent before bed. I woke up this morning and I was blown. Away. Watch the screencap of the scene it produced in the game. If you don't get blown away too, maybe i'll send you $20. It worked autonomously for a couple hours to produce this. Wait till you see my agent prompt in the 🧵⬇
5
1
1
7
639
Sogni World Kit is open-source btw! github.com/Sogni-AI/sogni-wo… Fork the skill, do something cool with it, and share! I'd love to see what made your brain explode too. Submit a pull request if you want to add something missing. If you've played around with it over the last couple weeks, make sure you tell your agent to update the skill because it's got some major updates over the last few days: Extensive story, dialogue, continuity and playback work. World Player bug fixes. Better slop detection, more nuanced transitions, support for more video models like Seedance 2.5. Pre-deployment transcoding of all media assets. Don't have a Claude or Codex agent? The skill should work with Hermes, OpenClaw, GrokBot etc. We even have a builder that runs directly from worlds.sogni.ai powered by Deepseek v4 Flash on the network. Finally, we currently have a creator contest with 3 months of Sogni Unlimited Pro and cash going to the top world creation submission. You can ask your agent to submit your world on your behalf and it'll populate a submission link for you: worlds.sogni.ai/submit-world…
3
72
The World Kit was trained to really leverage the 200+ open source models we have on Sogni Supernet render network extremely well. Models like Sam3 selection masks, Music models for soundtrack. Image to 3D models. With a $20 plan you can generate MiniMax H3 worlds to your hearts content and I have a ton of my own examples at worlds.sogni.ai But with this project I wanted to see how high we could get production value using the insanely powerful Seedance 2.5 Uncensored model which is also offered by Sogni at pay-as-you-go rates. I was floored at the polished output. But going through the agent scratch folder of failed generations had me crackin' up. Like these hallucination here. The agent reviewed the results, saw when things needed to be regenerated and did the work. But what's cooler is that it knew when clips were recoverable and it would recover them. Seedance 2.5 is an expensive model. So it would cut out any bad segments and regenerate them, and then restitch them with FFmpeg locally. And that is just the start. The techniques it would use to massage and collage the outputs blew me away again. We'd like to take credit for all of this nuanced creative intelligence but honestly a lot of it is the LLM itself. Claude Code's Opus 5.5 is just incredible. And if you steer that intelligence with specialized agent skills and tooling... well it's a brave new convenient world for creators now.
1
151
The prompt to my Claude Code Opus 5.5 (with Sogni Creative Agent kill + Sogni World Kit agent skill installed): Lets add one more pitstop in Sogni Meme World. 4chan memes in a desolate dirty place like Skid Row. A seedy place full of poverty drug deals, and prostitution. Here, Doomer Girl asks if Jean is looking for a good time. Here we meet Pepe the frog, crying, holding out his hand with a butteryfly on it, wondering why he's been branded a hate symbol. This scene world is a cast of 4chan / neckbeared meme culture like Wojak, maybe huffing paint. Carl from aqua teen hunger force drives by in a sus manner while they are chatting and tries to pick up our cohort. "Hey, how much for the pink guy, what kind of equipment it uh.. working with down there? I mean, it don't matter much to me" The scene could end with Jean having a heart to heart with Pepe who says his famous "Feels good man" line. It can feel very heartfelt and they hug it out, until Pepe whispers "So am I going down on you fellas, or are you guys going down on me? I mean I only got $20." His face changes to TrollFace, as PedoBear eases his face into the screeen pretty large (perspective) and our posse looks at each other shocked and scared, and all make a mad dash away, as they run they see a house on fire and inside we get a quick shot of This is Fine dog. The camera follows them as it drone shot flies higher and away as carl's voice yell's "Guys, i'm here for a good time, not for a long time, hurry back ok?" as the drone view flies out to reveal the full map ready for the user to select their next destination again. This is my general idea of a very funny, shocking pitstop. You get the general idea, you can massage this into a rapid fire funny scene. Since our platform is uncensored we're going for a pretty raunchy over the top "I can't believe they went there" kind of vibe. You don't have to include everything I wrote above, its just a general idea, see what you can come up with in this spirite. And don't shy away from vulgarity, or even a little nudity for shock value. You should use Seedance 2.5 uncensored as needed for these generations. I'm excited to see what you come up with. ---- It went to work, first generating character sheet references to use in the videos, using my Sogni Unlimited $20 plan.
🤖 Made with AI
2
263
Sogni World Kit just made the coolest interactive video web experiences as simple as chatting with your Agent. My vision: Sogni Meme World. impossibletriangle.io/experi… No coding, no prompting, just Agent execution straight from your dome to the page. Now with Seedance 2.5 Uncensored. How I built it 🧵
5
1
7
155
Step 4: Sign up for a Sogni Unlimited free trial Every free Sogni account comes with a 3-day free trial of Unlimited. Thats plenty of time to experiment with your first world for free. I'd love to see what you're building with Sogni Worlds Kit and the Sogni Creative Agent Skill, drop a comment and I'll check out your world! sogni.ai
1
27
Step 3: Into the weeds Want to know how it works behind the scenes? Here's a breakdown of the main models and brushstrokes your agent is working with: worlds.sogni.ai/how-this-was… Here's a writeup my Claude and Codex agents wrote about their experience building 2 of my worlds: blog.sogni.ai/blogs/how-agen… And behind the behind the scenes? The Sogni Creative Agent Skill: github.com/Sogni-AI/sogni-cr…
11
Step 2: Submit your world to worlds.sogni.ai/ There are plenty of great example worlds here: Bringing travel photos to life, creating a choose-your-own-adventure horror masterpiece, whatever you want. Sogni is uncensored so if you can dream it, your agent can build it. The coolest part: You can build multiple worlds with a single $20 Sogni Unlimited subscription. Build with Minimax H3 and 200+ other open-weight models. Premium vendor models like Seedance 2.5 are available too at a discounted pay-as-you-go rate for Sogni Unlimited subscribers. The agent and the models build the music and ambiance, the foley, the dialogue, the overarching storyline, the image and video assets, characters, the interaction points, the intuitive UI, the artful transitions, the cut scenes, the mobile optimized delivery--with your careful feedback or your casual feedback, with a review process that makes sure you land your vision. Sogni is running a contest, top Worlds get 3 months of Unlimited Pro + $300, 2 months of Unlimited Pro + $200 and 1 month of Unlimited Pro + $100. Deadline is November 8!
55
github.com/Sogni-AI/sogni-wo… Step 1: Install the Sogni Worlds Kit skill on your Claude, Codex, or Hermes. Then just type or voice note your Agent a few paragraphs of your vision and have it build interactive stories told in films, voices and music. Your agent gets access to Seedance 2.5 uncensored along with 200+ other models and the finesse to use them well.
37
Mark Ledford retweeted
MOSS-Image SFT and MOSS-Image 8-Step are now on Sogni, and both are included in Sogni Unlimited It’s OpenMOSS’s 5-billion-parameter text-to-image model, made for text. Quote the words in your prompt and it writes them out exactly, from long instructions and terminal commands to full menus, in English and Chinese. Every image in this thread was made with it 👇🧵
9
8
43
1,998
Mark Ledford retweeted
Layout Composer is here on Sogni ☕️ Instead of hoping a prompt puts things in the right spot, you place them yourself. Draw boxes for the cup, the cookie and the headline, and the image follows your layout. Works with MOSS-Image SFT, MOSS-Image 8-Step and FLUX 3 Image Klein 👇🧵
2
2
25
1,760
Mark Ledford retweeted
FLUX 3 Klein is now on Sogni, and it's included in Sogni Unlimited from $20/month. One model that creates and edits, fast: a 1024px image in under 2 seconds on Sogni's GPUs. Here's what you can make with it this autumn 👇🧵
12
17
1
58
2,716
I’ve worked on a lot of cool Creative AI product at Sogni but our recently released Sogni Worlds Kit takes the cake. I’ll share a nice 🧵 about it soon!
What if your next AI image was a place someone could explore? 🦥 Sogni World turns scenes into interactive adventures. Click a door to travel, meet a character, or discover an object you can collect. Every choice gives you another reason to look closer. This video follows Simon, our pink Slothicorn, through The Dream Thread: 23 places and 77 journeys, with routes that branch as you explore. Built on the Sogni Supernet with Krea 2, Krea 2 Identity Edit, SAM 3, MiniMax H3, MiniMax Music 3, Pixal3D and Qwen3-TTS. AI agents brought the pieces together using the Sogni Creative Agent skill. You can build from an idea or your own photographs: a strange story, a branching adventure, a trip you want someone else to experience. Here’s how the seven steps fit together🧵👇
3
51
Mark Ledford retweeted
Twin powers, activate! ⚡ Brothers @fishmongr and @krunkosaurus are landing in Singapore for TOKEN2049. Who’s ready to meet them? Oct 7 → luma.com/AI-You-Can-Touch Oct 9 → luma.com/dwfcywj8
8
4
27
1,348
I don’t leave the AI shipping dungeon often so come say hi 🍻
.@fishmongr is officially coming to Singapore for Token2049. Brother's gonna work it out. Token2049 are you ready?
1
1
73
During TOKEN2049 week, we’re hosting AI You Can Touch in Singapore. We’ll have 2 interconnected NVIDIA DGX Sparks running an open multimodal model live in the room - and everyone attending can connect over WiFi and try it themselves. We’ll dig into: • What you can realistically run locally today • The economics of owning vs renting AI • Why open-source + local AI are accelerating • What this shift means for founders, investors & builders Plus a fireside chat with Sogni AI, OpenBuilder and TZ APAC. 🍻 Food + drinks included 🕑 2–6 PM 📍 TZ APAC, Singapore 🎟️ Limited capacity — RSVP early
4
4
21
673
Mark Ledford retweeted
What an Nvidia GPU earns on Sogni Supernet in September 2026, before costs and running 24/7: RTX 5080: $174 to $242 RTX 4090: $259 to $357 RTX 5090: $413 to $594 RTX PRO 6000: $813 to $1,102 The low end is the current pool. The high end assumes the pool grows another 50% while your share of the work holds. Built on nine full days of worker activity, with video completions up 92% since Sept 1, most of it MiniMax H3. The worker results behind this forecast come from one carefully managed account. These observed results and forecasts are not evidence of guaranteed outcomes and do not establish what another account or GPU will earn. Actual earnings depend on keeping your machines online, sharing popular models that your hardware supports, keeping your equipment well cooled, and demand on any given day, week or month. Full breakdown 👇
1
4
28
1,792