There was a discussion in the Patron’s chat after the awesome Logik Live 164 on Sunday, about the next killer feature in Flame.
What do you think a killer feature would be? Please drop a comment and write your idea however simple or complicated it is. Take a look at other people’s suggestions and +1 them if you agree. Participation is key to steering Flame to a bright future.
Good questions. Here are my two entries.
There is a lot of energy around masking/matting/segmentation. The number of tools competing for attention in BorisFX, SmartRoto, ComfyUI, or even Flame feature requests that try to ease our burden with accurate and complex masking and matting to wit.
If someone took a step back and reinvisioned what the actual problem is that we’re trying to solve, and came up with a flexible tool that has pixel level accuracy in highly complex situations (movement, occlusions, lighting), can provide segmented selection, and much more - at a quality level like a 20-node Nuke tree for GS comp provides, that would be a game changer. We could work faster and better if our masking/matting problem can be considered ‘solved’.
I envision that not as a single ML model that does it all. We have seen that this doesn’t exist. But more like a tool that can employ traditional and new ML capabilities in a multi-step process to achieve the task with less human busy work or risk.
Maybe Modular Keyer meets AI agent + models + SmartRoto style robo roto.
The other one would be a translator of the existing VFX knowledge where an interactive conversation (like Claude chat) could build complex node tress and debug them, and provide solutions like additive keying, fine procedural hairline masking, and many more examples.
When you chat with Claude in Opus 5 you can have deep conversation about CV and compositing workflows with complex channel operations. It can read your tree and guide you into building solutions. There is tremendous knowledge available on all these techniques, but it’s not easily accessible or front-of-brain type of stuff. If there was a way that this knowledge could be made more accessible with an agent that helps you build your batch node tree in collaboration with you, that could make Flame the app every VFX artist would want to use.
This is not meant to put us out of work. You still need to understand the problem space, and what needs to be done. But it would make more advanced techniques more accessible on an everyday basis for working artists in an efficient manner. If you have downtime between projects (rare) you could work through this step by step and then save setups. But this would make some of these solutions buildable within project timelines.
That solution would be something like a special purpose MCP coupled with a pre-trained agent on CV and comp knowledge who can work with you directly in Flame and even evaluate and test results. And where possible it should take the results of these where allowed and feed back into the training corpus while protecting privacy of imagery we work with. All within a trusted environment maintained by ADSK.
I have a chat log with Claude that could illustrate this. Can share privately.
While I believe local models are the way for production environments.. API nodes that would allow piping into ‘the-next-best-video/image/omni/model-ever’ would be nice. Adsk could even setup a ‘credit’ system for using it, I wouldn’t mind if it’s competitively priced.
I would like to see an Ai generated camera track with points that also track objects. Basically feed in shot, get back fbx.
I think mine at the moment would be a temporally consistent ML based infill tool that is trained on 16/32bit HDR sources and can function well natively in both log and linear.
Better multi-channel EXR management a la nuke
This. Integrated at the timeline/batch/action/io levels as first class citizens. Essentially reimagining what a timeline is and what it can represent.
Yes. The amount of time I spend Che king and rechecking edits in this age of no edit lock is painful. I built a tool in chatgpt to help analyse stuff and speed up the process but I’m not allowed to use it. Fucking daft. It would help if it was built in.
The context was ADSK making a big bet on a feature that would stand out in the industry, that would be noted and other people would envy, sad that their platforms are now trailing. Of course in this day and age it only takes 9 months before BMD copies a shadow of your feature. But at least you got the sunshine and kudos for a bit, and proved that Flame still has it.
We all have a long wishlist, but some of them just create parity or play catch up, they wouldn’t put Flame into a pole position. We should think big and bold, even if it may never happen.
I can comp in 2d like 3 times faster than someone in Nuke by merit of the speed of our softwares alone. I would say that I would vastly prefer that autodesk throw all the ML and AI research time in the trash and devote that instead to making it so I can comp CG faster than Nuke too. I can sit a Nuke artist at my machine and blow them away by the speed, but the second I have a piece of CG to comp its a no deal.
This every day of the week. The 3D environment being faster and more user friendly and being able to utilise the multilayer passes within an exr sequence.
I’d add to that being able to create 3D meshes from point clouds as well as being able to unwrap & rewrap UVs from geo.
I can also see some incredible use cases for Gaussian splats to rebuild environments within Flame for cleanup work on 3D scenes or combining takes where there is a lot of parallax. I’d definitely place this request behind what @BrittCiampa has said though.
Last year I had to add a fullmoon to night time low lit shot. Nothing to track at all even with exposure and degtain right up. Flame could not do it and it wasn’t flames fault. I ran it though Runway and boom full moon tracked perfectly. So for me and AI tracking algorithm would be my No.1 killer feature within Flame.
Is the killer feature this?
More open integration with APIs like Runway etc?
I have been told that python integration in flame is much simpler than nuke by someone who has a lot of experience in nuke.
Weavy is doing this already (integrating APIs) but its UI is a bit clunky. I always have to import into flame to judge properly.
It’s possible to export to APIs like nano banana yourself with a bit of Python of course (and comfy) but if it were as easy as installing an OFX would that appeal?
Flame / Flare indie version with lower entry price so all can join in! 