Nicolas Vogel, Senior Flame Artiste and VFX Supervisor at Prodigious in France headed the Orange – la Compil des Bleues TV commercial which has gone viral the world over. One of the most impressive pieces of VFX for storytelling we have seen in years. And it is all the more impressive when you learn that all the imagery was just stock footage and most of the compositing was done without AI in Flame. We spoke to Nicolas in France about this incredibly effective piece of communication and equally impressive compositing.
fxguide: How many shots did you and the team do?
Nicolas: Initially, we were supposed to do 5 game actions, broken down into 3 shots. That’s 15 shots to do. But the agency asked us to fake as many shots as possible to make a very dynamic edit. In the end, we did 35 shots.
fxguide: The VFX was done in Flame, is that correct?
Nicolas: 90% of the shots were done using Flame. All the ‘face replacement’ work was done in Flame.
All the women’s hair was erased on the Flame, often frame by frame, using the ‘paint’ tool.
The motion design for the “Reveal” sequence was done on AfterFx, so we had to create a fake software interface to do the “Before/After” shots.
Contrary to what many believe, this is not a Deepfake solution. Nevertheless, I am curious to see what AI is capable of doing with these types of effects.
fxguide: Did you consider doing more of a deepfake/ ML solution? If so why did you favor Flame?
Nicolas: This project really began a year ago. The Marcel agency approached us with the idea of replacing female soccer players with players from the French national team. Quentin Martin, who was the Head of Podigious Post-production at the time, came to talk to me about the project to see if it was feasible. I immediately thought it was a brilliant idea: we absolutely had to succeed in bringing this project to fruition.
We immediately considered 3 different techniques:
- DeepdFake,
- 3D Mapping and
- compositing on Flame.
I’d worked on face replacement a few months earlier on other projects and had been able to compare these different techniques, whose limitations I was well aware of.
Deepfake takes a very long time to produce because you have to feed a lot of images into the ML. It’s also necessary to retouch the Deepfake renderings, as they don’t always work very well, especially with complicated movements. It was therefore risky to choose this technique.
The second option was 3D mapping. But here again, production times were too long. And a 3D scan of the players would have been necessary. Production costs were also too high.
Changing faces with Flame was a trick I’d already done, and didn’t think it was impossible to do, as long as you had the right image sources.
We suggested this technique to Marcel. They then asked us to carry out tests on 2 shots to see if we could come up with a sufficiently realistic result. A week later, the agency validated the tests.
fxguide: Can you outline the process in flame for a typical shot?
Nicolas: Each plan corresponded to a single Batch made up of several Action Nodes. Several graphic designers could then work on different parts of the shot at the same time.
First, we had to completely erase the players’ heads by recreating the background, i.e. the pitch or the stands with the spectators…
Then we rotoscoped the men’s heads, stabilized them and tracked the women’s movements. We also had to retiming the men’s images. Then we worked on the junction between the neck and the collar of the shirt, using warping (Distort tool).
Next, we worked on the player’s morphology:
– erase the chest (deformation work but also eliminate of shadows on the shirt).
– The arms, chest and thighs were enlarged.
– Muscles were redesigned ( Paint + Tracking ) and skin colors were adjusted ( Keyer + Roto + Color Warper ).
Finally, we had to change the color of the clothing, change the name and number on the shirt and add the 2 stars on the jersey and shorts where visible.
On all the wide shots, we erased the players’ long hair, often frame by frame. This was particularly tricky for actions behind the goal nets.
We also erased players when there were too many in the wide shots.
fxguide: Can you discuss how you found matching footage in the men’s footage for each shot? Was it from game footage or was it new footage shot just for this?
Nicolas: After my initial tests, I thought it would be important to organize a shoot with the male players. I wanted to shoot them on a spinner with several cameras, so as to get them from all angles. But it was impossible to get all the players together for a shoot, so the idea was quickly abandoned and we had to rig solely with footage from real matches; archive footage.
We had access to the FFF image bank. Les Artisans du Film” took care of the image search and made a pre-edit of all the best actions of the players in all the matches of the last 3 years. It took them 3 weeks. I then eliminated the shots that seemed too complicated to fake, and while an editor began the final cut, they started to search for the corresponding images in the men’s matches.
fxguide: How long did you work on the spot?
Nicolas: I did the 1st test a year ago.
Image research by “Les Artisans du Film” lasted 3 weeks.
Sébastien DELECOUR and Emilia Redondy, two Graphic Artists from Prodigious, worked with me for 6 weeks on Flame, rigging most of the shots.
During the last 2 weeks, many other graphic designers and assistant graphic designers from Prodigious then helped us finish on time.
In the end, it took 2 weeks of editing, 70 days of Flame, and 20 days of After FX.
The spot took 7 weeks to produce.
fxguide: How complex were the grading aspects to ’sell the composite’ – I assume lighting was critical?
Nicolas: Generally speaking, to compose 2 shots well, you need the same light direction, camera axis, and focal length. If you can achieve this, you’ve done 50% of the job.
By choosing to use only official matches, we’ve finally met these criteria.
If you’re used to watching soccer matches on TV, you’ll notice that the cameras are always in the same places: so we were pretty sure we’d find matching shots, in the same camera axes. The same applies to lighting: matches are often played in the evening, and stadiums are always lit in the same way, with light coming from all sides. Look at the players’ shadows on the pitch: there are always 4 shadows on the ground. The light matches almost every time.
I have to admit that I hadn’t anticipated this, and it was a really pleasant surprise. Sometimes, you also need a bit of luck!
fxguide: Were there many shots you looked at or started and then discounted or dropped – or was every shot it the original edit?
Nicolas: I eliminated from the selection all the shots that seemed too complicated to make: too long, too many players in the picture, and/or the movement impossible to match.
While searching for shots on the FFF image bank, my post-producer, Jonathan Keita, also found other shots, more interesting to fake. Other camera axes enabled me to cut out the action. So, yes, I was able to change shots in the montage when it suited me. On the other hand, we had agreed with the agency to absolutely rig certain key shots in the editing, such as Karchaoui’s 1st dribble, Le Sommer’s final celebration, and Cascarino’s incredible run.
fxguide: What resolution was this finished at? I assume the base women’s footage was HD?
Nicolas: Our image sources were HD (1920×1080), which is also the resolution of the final version.
fxguide: Is there any one shot that you were particularly proud of?
Nicolas: I’m pretty proud of the last shot because the 2 images match really well. We had a last-minute request for this shot: to add long sleeves to the shirt because Antoine Griezmann always plays with long sleeves. Here too, we used the Flame’s Motion Warp and did it in just a few hours.
But above all, I’m proud that we were able to rise to the challenge and help make this project possible.
It started with a fantastic idea, and we absolutely had to live up to this great concept.
Marcel agency always trusted us and never seemed to doubt us. I am truly grateful to them for giving us the opportunity to work on this project.
The spot created a buzz around the world, and although we had a feeling that people would appreciate it, we couldn’t have anticipated such a huge response.
I’m really happy for Antonin Jacquot and Vincent Teffene, from Marcel, who came up with the idea. This success is truly well-deserved.
fxguide: thanks so much and congratulations again