This tutorial demonstrates how to create professional AI football match videos by first generating a character reference sheet using GPT Image 2 to ensure consistent character appearance across all shots, then using the Flows Agent to generate multiple camera angles with specific style rules and audio directions, converting approved keyframes into video using Seedance 2.0, and finally adding broadcast commentary with Eleven v3 voice synthesis. The key workflow principles include maintaining character consistency through reference sheets, using structured prompts with style rules and camera angles, generating extra seconds per clip for editing flexibility, and adding audio post-production for professional commentary.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
Put Yourself in a Football Match with AI (Full Workflow)
Added:Wait, number 11? Is that Alec? He was not even on the team today. Alec has the ball. HE RUNS RIGHT PAST THE DEFENDERS.
HE SHOOTS. GOAL!
WHAT A GOAL BY ALEC. THIS IS CRAZY. The guy who just stepped in wins the cup.
>> So, that's me scoring a goal in an intense football match in a sold-out stadium with full broadcast commentary.
I made myself a custom kit, placed myself in unique shots, generated an action-packed sequence with a fully customizable football commentator voiceover, all inside of 11 creative.
Here's the full workflow to how you can place yourself in a football match using AI. The first thing you need is a high-quality picture of yourself. And with this picture, we're going to create a character reference sheet of you in your custom football kit. To do this, inside 11 creative, we're going to go ahead and click on flows. And here, we're going to click on new flow. And in this flow, we're going to create the entire football scene. And we're even going to be using the flows agent in just a second. But first, let's create the character reference sheet. To begin, the first thing I'm going to do is simply paste an image of myself. And so, you can upload any image of you, and remember you want this to be a high-quality image. Then, I'm going to drag out a connector from our reference image, and we're going to create an edit image node. And with this node, we're going to be using GPT image 2 to create the character reference sheet. So, we want to make sure that we're using GPT image 2. And I want to create with an aspect ratio of 16 by 9 because it's a good aspect ratio to create character reference sheets. Now, what I'm going to do is paste in the prompt, and we're going to put the prompt on screen for you so you can screenshot it, run it through an LLM, and then copy and paste that. But basically, this is what the prompt is doing. We're going to create a character reference sheet for a young male football player. We're going to be using, and then I'm going to tag the reference image. So, use this image right here as the face and identity reference. So, we want to preserve the exact face and details for character consistency throughout all of the generations of our AI football scenes.
After that, we're describing what the character reference sheet is, and it's essentially just multiple different shots of the same character with its face from the side, from the front, from the back, in all of the outfit that we're about to create. And so, here at the bottom is where you'll put in the description of your football kit.
Because we want our character, or our football player, to have the same number, the same colored kit throughout all of the generations. Otherwise, it won't look professional. And here to keep it simple, I've got the football kit is orange with the number 11 on the back and front long sleeves. But, again, you can describe any football kit that you'd like to have. And now, because this is a character reference sheet, we want to make sure that we increase the resolution. I'm going to set it to 4K, and then change the quality to high, and simply click run. And now, as you can see, we have my character reference sheet. And so, we've got me from the front, from the side, both sides, which is important, and then also from the back. So, no matter what scene we're generating, and no matter what model we use, the model is going to know what the character is supposed to look like from all angles. Because we're consistently going to be using this as a reference for all of the shots. And once we've got the character, we can move on to the fun part, and that's actually generating the scenes and turning those scenes into videos, and stitching it all together with commentary. And to do this, we're going to be using the Flows Agent. Now, everything we've done, upload the image reference and create the character reference sheet, we've done it manually.
And we can continue creating everything manually. But, when you're creating bigger projects like this, and you want consistency throughout the entire thing, well, you can save a lot of time by actually asking the Flows Agent to help you build it. And this is exactly what we're going to do. Now, the prompt you're seeing on screen right now is the prompt that we're going to use for the Flows Agent to create the entire football scene video. As you can see, at the top, we're saying create a three-shot football sequence of, and then here, we are tagging the reference sheet, so it knows the specific character that is taking the free kick.
Below that, we've got inputs. And so, here, we're using the character reference sheet, and so, once again, we tag the character reference sheet and the face reference that are already on the canvas. So, as you can see, we've got these two things right here. Well, in the Flows Agent chat, we are actually telling it what image to reference at what moment within the prompt. And after that, we've got the style section, which obviously we want to be consistent for all of the generations, which is why we've got repeat verbatim in every image prompt. This way, every time it generates a new scene or it generates a new shot, well, it's going to have the exact same style. And for this type of video, here it's very important that you customize the style at the beginning when using the Flows Agent. For example, here you could have more of a live broadcast one that looks like this, or you could have a Japanese anime style one that looks like this, or you could have more of a high-quality film production ad loop that looks just like this. And so, you could use one of those prompts, or you could actually customize your own. And then below that with the prompt, you've got the description of the three shots. And you'll notice what's important for the description of each shot is that we are showing the camera angle and the description of the actual image, but then we're also showing what that video should look like after we've gone from that image, which we'll likely use as a starting frame, to the actual video. And we're doing that for each of the shots. And doing it this way will allow Flows Agent to create the best possible images knowing what type of video it needs to be turned into afterwards, so it will adapt the prompt throughout the entire workflow, and it won't just think about that single generation. And finally at the bottom of prompt, we've got audio. And here we don't want any commentary and we don't want any music, we just want the noise of the actual stadium and the actual football, so the sound effects, because we're going to be adding in the commentary afterwards. That way we can have full control over how it sounds and it looks. Generating the commentary directly in the videos is tricky because sometimes you want to make a little bit of a change, but it means you have to regenerate the entire video. And so, we're going to do this afterwards with 11 V3, which has great voices for football commentary. And then finally at the bottom, we're just telling Flows agent the specs that we want. So, 16 by 9, and I like to generate 5 to 7 second clips. And the reason we do this is that these extra couple of seconds give us a little bit of leeway because obviously, we're not going to use the entire generation. We're going to stitch them together, maybe trim off a few seconds here or there, but it just gives us a little bit of room to play with. And then as you can see at the end, we like to end every prompt with consistent subject identity, natural motion, realistic ball physics, no flickering, no distortion, no unwanted text, no extra limbs. And this is just, you know, a little bit of negative prompting to ensure that we get consistent high-quality outputs. And once we've done this, in the Flows agent, we can now go and click send. And here, Flows agent is going to use all of this as context, and it's also going to use the reference sheet that we have and the initial face reference to help generate the best flow possible to create this football scene with me in it. And so as you can see right now, it's currently thinking. And if I zoom out a little bit, the Flows agent should then start building out the nodes it needs for these scenes. And so as we can see, the agent has built out the first step of our flow. And it's basically just saying it's going to generate three images that we can then approve before turning them into videos. And for each one, it's giving us the prompt, which we can preview. And so we can look through them, and you'll notice at the beginning of every prompt, we've got the little style prompt that we like. Then we've got the specific references that need to be used, and then it's got the actual description of the scene. And so we can go ahead and look through all of these.
If we're happy, we can simply click run all. If you weren't happy, you could just go into the chat and tell it what you would like to change. And now as we can see, the Flows agent has generated the three shots. The first one where he's standing over the ball in front of the goal with the wall of players in front of him, the second one where he's running up to kick it, and then the third one where he is celebrating. And you can also see that it's started to create the nodes with the video prompts.
And now here, if we're not happy with the shots, what we can do is we can go and customize the prompt and regenerate each of these shots before turning it into video. That way we're sure that we're getting the style, aesthetic, and look right before you actually create the video. So, it's important that you're happy with your shots, and you can go in and customize the prompt simply by typing and changing them right here, or you can actually ask, once again, the Flows agent to make the changes if there's a bunch of changes you would like. For example, if you wanted to make the entire kit pink or anything else. But, once you're happy with the shots, now, as you can see, so it's built out the video nodes, and it's telling us in the Flows agent chat the next step. So, we've got to review the frames, and once we're happy with all of the frames, you can see that the Flows agent is using each frame as a start frame, and it's then telling us that we can go and generate the videos. And so, here, by default, it is using of different video models to choose from, all of the best video models in the world, whether that's C Dance 2.0, Kling 3.0, or maybe V03, which is a much older model. But, we could then say in the Flows agent chat, "Use C Dance 2.0 to generate all of the videos." And C Dance 2.0 will cost you more credits, but it's one of the more realistic models, or you could go and use one that's a little bit cheaper, like Kling 3.0 Pro. And here, we simply then have to click send, and now the Flows agent should go and change the model selected in these three nodes, so I don't have to do it individually, and then hopefully it will run them, or it'll ask me if it's okay before actually running them. And as you can see, in the chat, it's telling me that they're now all C Dance 2.0. We can see that they're all C Dance 2.0, and so we can go and say, "Please generate," and click send. And now, it's going to generate the three video shots based on the start images that we've approved.
And now, as you can see, C Dance 2.0 has generated our three shots, one where I'm facing the wall and the players, the other one where I kick the ball, then the last one where I'm celebrating. And so, you can see, this is a lovely goal, and these look like lovely shots at a glance. And there's always a couple of issues because this is AI, and there are ways that you can fix this. For example, in the first and second shot, we can see that the football isn't exactly the same. And so, what could we do to fix this? Well, we could go and use a reference image of a football that we've created, and then use that in a reference in all of our shots. So, we could have fixed this at the image stage. So, we could have used a reference of a football, let's say we really like this one. We could then use this image as a reference for the following generation to make sure that it's the same football and paste that in the prompt. And there are also other issues. For example, with this shot right here, which is an incredible goal, as you can see at the beginning, the ball is rolling. And so, here we might want to go and adjust the prompt of this generation right here, and say the ball starts off static as a free kick.
However, with AI, as I mentioned earlier, I like to generate a few extra seconds here and there, because we can get creative in the edit. And so, what I'm going to do now for each of the generations, is I'm going to click the three dots, and we're going to click on save to assets. Here, I'm going to save it to a folder called football film, and click save. And I'm going to do the same thing with each of the generations. The reason I'm doing this is because now if I go back into studio, and I create a new blank project, a video project, here in my files, I can then open up football film, and I can add all of the generations to the timeline. So, if I click on the plus icon for each, we've now got the three generations that I can put in the order of appearance in the timeline. And so, now I've dragged them in order, I've got them in order of preparing for the free kick, taking the shot, and then celebrating. And now, this clip is currently 18 seconds long.
But, what I want to do is I want to trim a little bit on each. And so, to fix this without regenerating, I can actually just trim from here onwards, so we don't actually see the mistake with the first football. I can then bring this back, and maybe I don't want to see it moving, and so maybe we start it from this shot right here. Drag it back, and then I might cut off just a little bit, so we see that it goes into the goal, just like so. And then maybe here we bring it back a little further where it goes up from the 11 on the jersey all the way to everyone crowding him and surrounding him. And so now if we play this from the beginning, let's see what it looks like.
Excellent. So now the audio cuts were quite harsh. So what I could do is I could actually move these to different tracks and then drag them out and then click on the clip and simply add a fade out. And now as you can see the audio is actually fading out while we already see the new clip and then I might fade in the new one just like so. And then we're going to do the same thing with the beginning right here. And so once again, I'm actually going to fade this in and then we might click on this one and fade this one out ever so slightly. And let me just bring it down onto the same track. So now we've got our three clips.
Now to add a voiceover, what I'm going to do is I'm going to click on speech.
Here, I want to go ahead and select 11 V3. With 11 V3, I can then go and choose a voice that works with 11 V3 and I can also add audio tags. And audio tags are essentially keywords that direct the delivery of the voice. Inside 11 Labs, inside the voice library, you've got multiple different commentator style voices and one of my favorites is the British football announcer. And so if I select this, now I can type in any text that I want the commentator to say and generate it and overlay it with this video. And so here, with the style of video I have, I'm imagining this is a little bit of a replay of a goal that we just saw because I generated it in slow motion. And so I'm adding commentary that will kind of fit this vibe. And so I've written this out and now what I want to do is simply click generate.
>> What a shot right there. You don't see that every day. Not surprised they made it to the finals with this man.
>> Excellent. And so now all I have to do is actually just line it up. And actually, I might adjust and end up just deleting this entire clip at the beginning, but I can go back to the flow earlier and regenerate some more, add them to the same folder, so I could drag them back into this project. And what I really like with flows is that let's say I've generated the perfect voiceover because I can click on it and actually regenerate as many times as I like, as many times as I need to. And I'm actually not a huge fan of the voice that's been generated, so I can actually customize the text or simply just regenerate it to get a new version to see what it sounds like.
>> WHAT A SHOT RIGHT THERE. You don't SEE THAT EVERY DAY. NOT SURPRISED THEY MADE IT TO THE FINALS with this >> And now what I want to do is actually adjust the timing. I like the timing of the first part, but I'm going to place the playhead and I'm simply going to split the voiceover so I can move this back a little bit further. So now if I play it, >> WHAT A SHOT RIGHT THERE.
YOU DON'T SEE THAT EVERY DAY. NOT SURPRISED THEY MADE IT TO THE finals with this man.
>> As you can see, the timing now matches up a lot better. And here I'm just showing you the basics. I used the first generation of everything I got. With AI, you usually need to regenerate a few times so you can get something that you can actually use. And so if I took an extra few minutes per generation, the final result would have looked even better. And for fun, I've actually gone and polished this off for you just by spending an extra 20 minutes, and this is what the actual final result looked like.
>> Wait, number 11? Is that Alec? He was not even on the team today. Alec has the ball. HE RUNS RIGHT PAST THE DEFENDERS.
HE SHOOTS. GOAL!
WHAT A GOAL BY ALEC. THIS IS CRAZY. THE GUY WHO JUST stepped in wins the cup.
>> And that's how to place yourself in a football match using AI inside of Lemon Creative. If you have any questions, drop them in the comments section down below. And if you enjoyed this video, please hit that like button and don't forget to subscribe. Thanks for watching.
Related Videos

Expanding Stikbot thumbnails
leopoldshorts
2K views•2023-09-24

Digital Discrimination: Cognitive Bias in Machine Learning
redmonktechevents2974
4K views•2019-12-18

Evolutionary Approach to Clustering by Ujjwal Maulik
ICTStalks
279 views•2019-06-26

Rose Yu "Learning from Large-Scale Spatiotemporal Data"
networkscienceinstitute
2K views•2019-03-04

Stanford Seminar - Generalization through Task Representations with Foundation Models
stanfordonline
4K views•2025-07-14

Satellite-Based Wheat Yield Forecasting using GEE & Transformer Neural Network
gisrsinstitute
634 views•2025-06-15

Paradigm Shifts in Data Processing for the Generative AI Era: Robert Nishihara of Anyscale & Ray.io
GradientFlow
2K views•2025-01-02

How to Build Your Own GenAI-Based Knowledge Management System
2150GmbH
360 views•2025-06-03
Trending

WOW! Judge TURNS THE TABLES on Trump in His OWN $10B LAWSUIT!!!
MeidasTouch
197K views•2026-07-23

Playstation NO DISC/NO BUY Fight Is Over...
DavidJaffeGames
4K views•2026-07-23

Steam and Xbox Just Dropped The Hammer On PlayStation
OhNoItsAlexx
9K views•2026-07-23

Americans Confused in Australia for 17 Minutes Straight
IWrocker
17K views•2026-07-23