Fix It In Prompt

AI agents can build the workflow, critique the result, and keep iterating, but when the service goes down, suddenly reaching for the mouse feels strangely primitive.This week on Fix It In Prompt, Alex Villabon and Doug Hogan talk about becoming dependent on agentic tools, keeping strong local models as production insurance, and finding the human checkpoints that should stay in the loop as AI moves deeper into VFX.We get into Qwen3.8-Flash-Next, a locally generated MiniMax H3 hotel-lobby meme, Seedance 2.5 Draft Mode, Seed Audio 1.5, Qwen-Image 2.1, LTX-2.5, Claude Opus 5.5, Jev’s fast decision model, pop-up VFX teams, Comfy Agent, Comfy MCP, Comfy Router, and the slightly uncomfortable realization that sometimes the artist is the bottleneck.💬 Where do you draw the line between letting an agent run and staying close enough to understand and improve the work? 📬 CONTACTEmail us: [email protected] Doug:ComfyUI & VFX website: https://www.doughogan.comComfyUI Workflows: https://comfyui.sjv.io/OYydOGLinkedIn: https://www.linkedin.com/in/doughoganFind Alex:VFX + AI newsletter: https://alexvillabon.substack.com/LinkedIn: https://www.linkedin.com/in/avillabon/🔗 LINKS MENTIONED IN THE EPISODEQwen3.8-Flash-Next — the local mixture-of-experts model Doug recommends:https://huggingface.co/Qwen/Qwen3.8-Flash-NextMiniMax H3 commercial licensing for local production work:https://blog.comfy.org/p/comfy-is-now-the-only-official-resellerQwen-Image 2.1 in ComfyUI:https://blog.comfy.org/p/qwen-image-21-in-comfyui-open-weightSeedance 2.5 — official model overview:https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5Claude Opus 5.5:https://www.anthropic.com/claude-opus-5-5Dario Amodei — “We Must Pace the Frontier”:https://darioamodei.com/post/we-must-pace-the-frontierTypeSafe — Introducing System One Models and Jev:https://typesafe.ai/blog/introducing-system-one-models-and-jevOstris AI Toolkit — open-source LoRA and model fine-tuning:https://github.com/ostris/ai-toolkitComfy MCP — local agent control for ComfyUI:https://blog.comfy.org/p/open-sourcing-comfy-mcp-on-localComfy Router — one API for frontier media models:https://blog.comfy.org/p/introducing-comfy-router-one-api⏱️ TIMECODES00:00 When the agent goes down and the mouse feels strange02:50 Local LLMs as production insurance05:53 Rebuilding the hotel-lobby meme with MiniMax H310:25 Learning prompt structure from agent feedback loops13:00 ByteDance, Seedance 2.5 Draft Mode, and Seed Audio 1.516:20 Pop-up VFX teams and the changing studio model21:39 Astra, Opus 5.5, and agents unlocking old software26:35 Pacing the frontier and protecting human checkpoints35:35 Open source, trust, and who benefits from slowing down38:44 Jev: fast decisions instead of generated text43:43 Learning through play and pushing past model assumptions47:42 Full access, autocomplete habits, and AI hygiene50:39 Junior artists, agent dependence, and the future of work59:29 Reverse-engineering agents, Comfy Agent, and Comfy Router#DellProPrecision #DellProMax #DellTech #NVIDIA #AI #VFX #ComfyUI #AIAgents #OpenSourceAI #MiniMaxH3 #Seedance #Jev #FutureOfWork

AI VFX is getting more capable by the week, but Victor Perez makes the case that the real job is still curiosity, taste, and knowing what decision to make next.This week on Fix It In Prompt, Alex Villabon and Doug Hogan are joined by VFX supervisor, director, and educator Victor Perez. They talk through Astra’s growing Blender capabilities, why artists should treat new tools as a creative paradigm shift rather than a magic button, and how the same playful experimentation that built a career in Nuke still matters now.We get into Victor’s new AI for Film and VFX book, open versus closed models, workstation realities, DiffHDR’s LDR-to-HDR workflow, GPT Images 2.5’s more surgical edits, and the wonderfully janky Astra-in-Blender Rickroll. Along the way: pixels, proxies, heat, community, and why “Will Smith eating spaghetti” may have a new 3D benchmark.💬 Where are you drawing the line between a stable production workflow and the newest AI experiment? Let us know in the comments.📬 CONTACTEmail us: [email protected] Doug:ComfyUI & VFX website: https://www.doughogan.comComfyUI Workflows: https://comfyui.sjv.io/OYydOGLinkedIn: https://www.linkedin.com/in/doughoganFind Alex:VFX + AI newsletter: https://alexvillabon.substack.com/LinkedIn: https://www.linkedin.com/in/avillabon/🔗 LINKS MENTIONED IN THE EPISODEOpenAI — Introducing ChatGPT Images 2.5:https://openai.com/index/introducing-chatgpt-images-2-5/Eyeline Labs, Netflix, and Texas A&M — DiffHDR: Re-Exposing LDR Videos with Video Diffusion Models:https://eyeline-labs.github.io/DiffHDR/Victor Perez — Generative AI for Film and VFX: From Creative Concepts to Studio-Ready Workflows:https://www.routledge.com/Generative-AI-for-Film-and-VFX-From-Creative-Concepts-to-Studio-Ready-Workflows/Perez/p/book/9781041235927GPT-6 Astra Rickroll recreated in Blender:https://www.reddit.com/r/singularity/comments/1w9aeyk/gpt6_astra_rickroll_in_blender/Will Smith eating spaghetti, made with Astra in Blender:https://www.instagram.com/reels/DdABYePTpgK/⏱️ TIMECODES00:00 Welcome Victor Perez02:47 Astra changes the conversation07:00 Curiosity beats fear10:01 Directing the tool instead of pressing a magic button13:20 The Shake-to-Nuke transition—and learning the next tool19:13 Artist first, technician second22:02 Astra makes a ’90s sitcom starring Doug’s dog27:03 Open weights, trust, and stable workflows32:01 Victor’s production-minded AI VFX book39:14 Future-proof ideas and AI-assisted editing48:12 DiffHDR and inventing HDR data from LDR footage57:15 GPT Images 2.5, precise edits, and Astra demand01:05:04 Breaking realism to find new stories01:10:33 Astra, Blender, Rickroll, and the next spaghetti benchmark#DellProPrecision #DellProMax #DellTech #NVIDIA #AI #VFX #ComfyUI #Blender #Nuke #ChatGPTImages #DiffHDR #GenerativeAI #Filmmaking #VisualEffects

GPT-6 Astra is not a one-click VFX solution, but it may be the first tool that makes a lot of specialists ask what else they could build.This week on Fix It In Prompt, Alex Villabon and Doug Hogan hit the bat signal for an emergency episode about GPT-6 Astra and what its computer-use abilities mean for VFX, 3D, and creative work. We talk about NVIDIA’s acquisition of Hugging Face, World Labs’ Atlas world model, and the difference between watching a demo and actually handing a model a task.Then we get into Astra building and revising Blender scenes, Godot, guiding ComfyUI workflows, blocking shots for Seedance, matching a DaVinci grade, and turning a vague submarine-game idea into something Doug could play with his son. The useful part is not that the first pass is perfect. It is that artists can bring taste, notes, context, and production judgment to tools that can now do much more of the execution.💬 What have you actually handed to a computer-use model—and where did you still need to provide taste, notes, or production judgment?📬 CONTACTEmail us: [email protected] Doug:ComfyUI & VFX website: https://www.doughogan.comComfyUI Workflows: https://comfyui.sjv.io/OYydOGLinkedIn: https://www.linkedin.com/in/doughoganFind Alex:VFX + AI newsletter: https://alexvillabon.substack.com/LinkedIn: https://www.linkedin.com/in/avillabon/🔗 LINKS MENTIONED IN THE EPISODEWorld Labs — Atlas: A World Model for Spatial Intelligence:https://www.worldlabs.ai/blog/atlasOpenAI — GPT-6 Astra: A New Generation of Intelligence:https://openai.com/index/gpt-6-astra/Reddit — Fable 5.1 vs GPT-6 Astra in Blender:https://www.reddit.com/r/OpenAI/comments/1w7ppcj/fable_51_vs_gpt_6_astra_3d_blender_mind_blowing/Reddit — GPT-6 Astra building a 3D game with Blender MCP:https://www.reddit.com/r/accelerate/comments/1w7nzal/dude_gpt6_astra_is_some_kind_of_turboagi_machine/@TopviewAIhq — GPT-6 Astra / creative workflow post:https://x.com/TopviewAIhq/status/2096140020340191251@petergostev — GPT-6 Astra / 3D workflow post:https://x.com/petergostev/status/2096264296099459079@AiBattle_ — GPT-6 Astra / creative AI post:https://x.com/AiBattle_/status/2095994051354919049@luccacerf — GPT-6 Astra / 3D workflow post:https://x.com/luccacerf/status/2096308567863079420@higgsfield_ai — Astra, Seedance, and VFX workflow post:https://x.com/higgsfield_ai/status/2095622423324590308@wizardbrainz — GPT-6 Astra / creative coding post:https://x.com/wizardbrainz/status/2096151812973732312@sharifshameem — GPT-6 Astra / software-building post:https://x.com/sharifshameem/status/2095653641164329143@annalea_l — GPT-6 Astra / creative workflow post:https://x.com/annalea_l/status/2096009976103801082@Dimillian — GPT-6 Astra / app-building post:https://x.com/Dimillian/status/2095596700815516004@badxstudio — GPT-6 Astra / 3D workflow post:https://x.com/badxstudio/status/2095982983379653113@higgsfield_ai — Astra / VFX workflow post:https://x.com/higgsfield_ai/status/2096334648016253339⏱️ TIMECODES00:00 Emergency episode: Astra changes the conversation02:02 LTX-2.5, open source, and the VFX mind06:41 NVIDIA acquires Hugging Face15:55 World Labs Atlas and controllable 3D worlds21:24 What does AGI mean for VFX artists?30:33 From computer use to Blender, Godot, and 3D39:21 Astra, ComfyUI, and workflows as building blocks42:35 Grayboxing and guiding Seedance generations45:34 DaVinci Resolve, 3D assets, and a train-yard test53:30 AI as leverage for artists and small teams57:55 Building a submarine game with a son and a vague idea1:00:48 Don’t wait six months to start learning1:09:50 Taste, skills, and the artist’s judgment1:14:05 Getting time back—and staying curious#DellProPrecision #DellProMax #DellTech #NVIDIA #AI #VFX #GPT6 #Astra #HuggingFace #WorldLabs #Atlas #Blender #ComfyUI #Seedance #GameDevelopment #GenerativeAI

This week on Fix It In Prompt, Alex Villabon and Doug Hogan are joined by LTX VP Applied Research Ofir Bibi and Technical Director Mohamed Oumoumad to unpack LTX-2.5: how better data and Gemma 4 improve prompt understanding, why older LoRAs can survive model upgrades, and how Diffusion Fidelity Rendering spends compute where the final pixels actually matter.We get into open weights, LTX Trainer, IC-LoRAs, world models, real-time and 360 generation, alpha and depth outputs, multi-pass EXRs, physical AI, and why production teams need repeatable controls instead of a one-prompt movie button.💬 If you could pull one production pass or control out of an AI video model today—alpha, depth, motion, normals, clean plate, or pose—which would change your pipeline most?📬 CONTACTEmail us: [email protected] Doug:ComfyUI & VFX website: https://www.doughogan.comComfyUI Workflows: https://comfyui.sjv.io/OYydOGLinkedIn: / doughogan Find Alex:VFX + AI newsletter: https://alexvillabon.substack.com/LinkedIn: / avillabon 🔗 LINKS MENTIONED IN THE EPISODELTX-2.5 open-source overview:https://docs.ltx.io/open-source-model...Run LTX-2.5 with ComfyUI or Python:https://docs.ltx.io/open-source-model...LTX Trainer overview:https://docs.ltx.io/open-source-model...LTX LoRA and IC-LoRA guide:https://docs.ltx.io/open-source-model...ComfyUI-LTXVideo nodes and workflows:https://github.com/Lightricks/ComfyUI...⏱️ TIMECODES00:00 Meet the LTX team: Ofir Bibi and Mohamed Oumoumad01:39 LTX-2.5 closes the speed-versus-quality gap04:49 Data is king: captions, filtering, and reinforcement learning09:15 Better prompting can mean saying less13:07 LoRA compatibility across LTX generations16:33 What makes LTX a world model?20:42 Diffusion Fidelity Rendering: spend compute where it matters32:21 Mohamed’s journey from VFX artist to model trainer41:00 Open weights, community, and the LTX ethos44:43 One base model cannot fit every production49:02 Real-time diffusion, 360 video, and the road to LTX 5.054:44 Can a world model output VFX passes?01:04:44 Open models versus building production on moving sand01:07:17 Why VFX studios should fine-tune their own models#DellProPrecision #DellProMax #DellTech #NVIDIA #AI #VFX #LTX #LTX25 #ComfyUI #OpenSourceAI #LoRA #AIVideo #Filmmaking

There are too many good AI models to test, and that may be the best problem VFX artists have had in a while.This week on Fix It In Prompt, Alex Villabon and Doug Hogan try to keep up with FLUX 3, Seedance 2.5, MiniMax H3, LTX 2.5, MiniMax Music 3, and Qwen3.8 27B while asking what agents actually change in a production workflow. They get into MCP, ComfyUI, Blender, Hermes, Unsloth Desktop, local speech-to-text with Handy, and why an agent is more useful as a collaborator with tools than as a vending machine.Then Doug shows MiniMax H3 video-to-video, face-replacement, character-placement, and background-restyling experiments using masks, audio-driven performance, reference images, and LLM-built prompts. Alex shows how to let an agent generate overnight tests from a prompt guide. The through-line is simple: use open weights to mix models, make failures visible, and let artists curate more iterations.💬 What would you hand to an agent first: troubleshooting ComfyUI, building prompts, or generating overnight tests?📬 CONTACTEmail us: [email protected] Doug:ComfyUI & VFX website: https://www.doughogan.comComfyUI Workflows: https://comfyui.sjv.io/OYydOGLinkedIn: https://www.linkedin.com/in/doughoganFind Alex:VFX + AI newsletter: https://alexvillabon.substack.com/LinkedIn: https://www.linkedin.com/in/avillabon/🔗 LINKS MENTIONED IN THE EPISODEMiniMax Music 3 — open-weights music model:https://www.minimax.io/blog/minimax-music-3-0-next-generation-open-weights-production-ready-versatile-music-modelQwen3.8 27B:https://huggingface.co/Qwen/Qwen3.8-27BUnsloth Desktop — local AI workspace for models, tools, and training:https://unsloth.ai/docs/desktopHiggsfield AI Film Production Pipeline — production guide:https://drive.google.com/file/d/1MRQoNq7dJD3s1OGWi94BKxoTIac4w_33/viewHiggsfield Zephyr: Special — project and production assets:https://higgsfield.ai/@higgsfield.studio/projects/zephyr-specialHandy — open-source, offline speech-to-text:https://handy.computer/MiniMax Physics Test:https://www.reddit.com/r/StableDiffusion/comments/1vjxn9l/minimax_h3_physics_test_seedvr2_rtx_vsr/Alex's Random TV Prompter for MiniMax H3:https://www.youtube.com/watch?v=UHmmsGG1dg4LTX 2.5:https://ltx.io/model/ltx-2-5⏱️ TIMECODES00:00 Too many releases, too little sleep02:08 Why FLUX 3 and MiniMax are fun07:13 The model-release firehose09:52 Agents, Comfy MCP, and a one-minute generation12:01 What an agent actually is14:54 Unsloth Desktop as a local AI workstation17:15 Turning apps agent-centric with MCP19:03 Finding repetitive work worth automating24:42 The Higgsfield production guide and open-source project assets30:08 Handy, WhisperFlow, and local speech-to-text32:24 MiniMax H3 and the 32B text encoder37:04 Face replacement, audio-driven performance, and masking40:59 Open weights, model combinations, and community-built workflows44:17 Video restyling with LLM-built prompts48:08 Overnight experiments and what models can do50:07 From RFPs to overnight concept exploration52:14 LTX 2.5 and choosing the right model for the shot55:38 Experiment, share, and stay curious#DellProPrecision #DellProMax #DellTech #NVIDIA #AI #VFX #ComfyUI #MiniMaxH3 #LTX #FLUX3 #Qwen #OpenWeights #Agents #MCP #Filmmaking #VisualEffects