Logo for a startup with zero budget: constructors and templates
What a startup encounters first of all
Usually, a person opens one of the constructors, enters the company name, and gets a dozen logo options in a minute. It seems like the task is solved. In reality, only the first part is solved — the visual image has appeared. Then come the questions that for some reason are not discussed in the advertising descriptions of the services: in what format to download the file, whether it is legal to use this logo, and why the company name in Cyrillic sometimes turns into unreadable gibberish of letters.
Constructor or neural network: what to choose
There is no universal answer, but there is a working criterion: the more important result manageability is for you, the more sense it makes to use a specialized constructor rather than a pure AI generator.
Midjourney and DALL-E create an image from scratch, literally drawing pixels according to a text prompt. This is beautiful but unpredictable — getting a logo with a specific font, exact proportions, and correct text is almost impossible. Turbologo, Looka, or Flyvi work differently: they use ready-made icon libraries and typography systems to compose the result. The outcome may be less exciting, but it is manageable and suitable for real-world use.
| Criterion | AI generator (Midjourney, DALL-E) | Constructor (Turbologo, Canva, Looka) |
|---|---|---|
| Predictability of the result | Low, each time a new composition | High, works with templates |
| Vector file as output | No, only raster | Yes, SVG/PDF included in the package |
| Working with Cyrillic | Often distorts letters | Uses system fonts, text is readable |
| Price | Free or subscription to the service | Free preview, HD package ~990–3490 ₽ |
| Suitable for trademark registration | Doubtful without further editing | Yes, with minimal manual correction |
Practical conclusion: for ideas and inspiration, you can experiment with prompts in neural networks, but for a real business, it's better to assemble the final logo in a constructor or manually add details in Figma.
Why neural networks mangle Cyrillic
Language models that generate images are mainly trained on English-language data. Latin is a familiar sequence of characters for them, while Cyrillic is a rare guest, so letters 'float': they get mixed up, duplicated, and turn into pseudo-graphics. This is not a bug of a specific service, but a systemic limitation of all generative models as of today.
The simple and reliable workaround: don't ask the neural network to draw text at all. Generate only the graphic core — an icon, symbol, or abstract form. Type the company name separately, manually, in Canva or Figma using any free font with Cyrillic support. Combine the icon and text into one composition — and you get a working logo without distorted letters.
Raster or vector: the difference that decides everything for printing
Any free generator by default outputs a raster image — usually sized up to 1024×1024 px. This is sufficient for screens: websites, social media, presentations look normal. The problem begins with printing.
- A minimum of 300 dpi is required for business cards — a raster file 1024 px at this resolution turns into a logo the size of a postage stamp, and beyond that, it disperses into pixels.
- For a meter-wide banner, 100–150 dpi is sufficient, but even here, the raster from the generator often falls short of the actual size.
- Vector (SVG, PDF, EPS) does not have a fixed resolution — it can be scaled up to billboard size without loss of quality, because it is not a pixel grid, but mathematical curves.
The conclusion is simple: raster is sufficient for the web, and only vector is suitable for any printing. If a service does not provide SVG or PDF in a free tariff, this is a signal that to work properly, you will either have to pay extra or redraw the logo yourself.
What files need to be prepared for output
One file is not a finished result, but a semifinished product. To make a logo actually work in all contact points with the brand, a minimum such set is needed:
- SVG — a vector file for the website and any scaling without loss of quality.
- PNG with a transparent background — for social networks, presentations, documents.
- PDF or EPS in the CMYK color model — for sending to a print shop.
- A separate AI or PDF/X-1a file with the text converted to curves, without gradients, shadows, and transparency — a requirement of most print shops for printing business cards and banners.
An important note: a logo downloaded from the generator in RGB mode will likely be reverted for rework by the print shop. RGB is a color model for screens, CMYK is for printed inks, and without color conversion the output will look different from the monitor.
Copyright: what is allowed and what is risky
According to the terms of most services — Canva, Turbologo, Dynadot — a ready-made logo can be used for commercial purposes without restrictions. This is not a myth, it is stipulated in the license. The problem arises not with usage, but with trademark registration.
Rospatent requires that the object of protection be the result of the author's creative work, not fully generated by a machine without human input. A logo taken from a pure generator without any edits formally falls into the risk zone — the examination may doubt the existence of the copyright object at all. The solution is workable and simple: use AI or a constructor to get an idea and basic form, and do the final vector drawing with meaningful edits — adjust proportions, change line thickness, refine details by hand in Figma or Illustrator. This is already your creative work, and such a logo can be registered without problems.
Frequent mistakes made by startups with zero budget
Most often, money and time are lost not in choosing a service, but in ignoring technical details.
- People only download PNGs for the web and then try to scale them up for a banner — they get a blurry image.
- They order business card printing in an RGB file instead of CMYK and get colors that don't match their expectations.
- They ask an AI to draw a logo with the full company name in Russian and are surprised that the text is illegible.
- They register a trademark for a logo without any manual adjustments and get a rejection or a long back-and-forth with the examination office.
- They don't save the original vector file, limiting themselves to a screenshot of the finished image — and without the vector, any future edits become a complete redraw from scratch.
Zero budget does not mean zero quality — it means that part of the designer's work will have to be done by yourself: to figure out the formats, bring the file to condition manually, and think one step ahead, for what the logo will be needed in six months.
What a startup encounters first of all
Usually, a person opens one of the constructors, enters the company name, and gets a dozen logo options in a minute. It seems like the task is solved. In reality, only the first part is solved — the visual image has appeared. Then come the questions that for some reason are not discussed in the advertising descriptions of the services: in what format to download the file, whether it is legal to use this logo, and why the company name in Cyrillic sometimes turns into unreadable gibberish of letters.
Constructor or neural network: what to choose
There is no universal answer, but there is a working criterion: the more important result manageability is for you, the more sense it makes to use a specialized constructor rather than a pure AI generator.
Midjourney and DALL-E create an image from scratch, literally drawing pixels according to a text prompt. This is beautiful but unpredictable — getting a logo with a specific font, exact proportions, and correct text is almost impossible. Turbologo, Looka, or Flyvi work differently: they use ready-made icon libraries and typography systems to compose the result. The outcome may be less exciting, but it is manageable and suitable for real-world use.
| Criterion | AI generator (Midjourney, DALL-E) | Constructor (Turbologo, Canva, Looka) |
|---|---|---|
| Predictability of the result | Low, each time a new composition | High, works with templates |
| Vector file as output | No, only raster | Yes, SVG/PDF included in the package |
| Working with Cyrillic | Often distorts letters | Uses system fonts, text is readable |
| Price | Free or subscription to the service | Free preview, HD package ~990–3490 ₽ |
| Suitable for trademark registration | Doubtful without further editing | Yes, with minimal manual correction |
Practical conclusion: for ideas and inspiration, you can experiment with prompts in neural networks, but for a real business, it's better to assemble the final logo in a constructor or manually add details in Figma.
Why neural networks mangle Cyrillic
Language models that generate images are mainly trained on English-language data. Latin is a familiar sequence of characters for them, while Cyrillic is a rare guest, so letters 'float': they get mixed up, duplicated, and turn into pseudo-graphics. This is not a bug of a specific service, but a systemic limitation of all generative models as of today.
The simple and reliable workaround: don't ask the neural network to draw text at all. Generate only the graphic core — an icon, symbol, or abstract form. Type the company name separately, manually, in Canva or Figma using any free font with Cyrillic support. Combine the icon and text into one composition — and you get a working logo without distorted letters.
Raster or vector: the difference that decides everything for printing
Any free generator by default outputs a raster image — usually sized up to 1024×1024 px. This is sufficient for screens: websites, social media, presentations look normal. The problem begins with printing.
- A minimum of 300 dpi is required for business cards — a raster file 1024 px at this resolution turns into a logo the size of a postage stamp, and beyond that, it disperses into pixels.
- For a meter-wide banner, 100–150 dpi is sufficient, but even here, the raster from the generator often falls short of the actual size.
- Vector (SVG, PDF, EPS) does not have a fixed resolution — it can be scaled up to billboard size without loss of quality, because it is not a pixel grid, but mathematical curves.
The conclusion is simple: raster is sufficient for the web, and only vector is suitable for any printing. If a service does not provide SVG or PDF in a free tariff, this is a signal that to work properly, you will either have to pay extra or redraw the logo yourself.
What files need to be prepared for output
One file is not a finished result, but a semifinished product. To make a logo actually work in all contact points with the brand, a minimum such set is needed:
- SVG — a vector file for the website and any scaling without loss of quality.
- PNG with a transparent background — for social networks, presentations, documents.
- PDF or EPS in the CMYK color model — for sending to a print shop.
- A separate AI or PDF/X-1a file with the text converted to curves, without gradients, shadows, and transparency — a requirement of most print shops for printing business cards and banners.
An important note: a logo downloaded from the generator in RGB mode will likely be reverted for rework by the print shop. RGB is a color model for screens, CMYK is for printed inks, and without color conversion the output will look different from the monitor.
Copyright: what is allowed and what is risky
According to the terms of most services — Canva, Turbologo, Dynadot — a ready-made logo can be used for commercial purposes without restrictions. This is not a myth, it is stipulated in the license. The problem arises not with usage, but with trademark registration.
Rospatent requires that the object of protection be the result of the author's creative work, not fully generated by a machine without human input. A logo taken from a pure generator without any edits formally falls into the risk zone — the examination may doubt the existence of the copyright object at all. The solution is workable and simple: use AI or a constructor to get an idea and basic form, and do the final vector drawing with meaningful edits — adjust proportions, change line thickness, refine details by hand in Figma or Illustrator. This is already your creative work, and such a logo can be registered without problems.
Frequent mistakes made by startups with zero budget
Most often, money and time are lost not in choosing a service, but in ignoring technical details.
- People only download PNGs for the web and then try to scale them up for a banner — they get a blurry image.
- They order business card printing in an RGB file instead of CMYK and get colors that don't match their expectations.
- They ask an AI to draw a logo with the full company name in Russian and are surprised that the text is illegible.
- They register a trademark for a logo without any manual adjustments and get a rejection or a long back-and-forth with the examination office.
- They don't save the original vector file, limiting themselves to a screenshot of the finished image — and without the vector, any future edits become a complete redraw from scratch.
Zero budget does not mean zero quality — it means that part of the designer's work will have to be done by yourself: to figure out the formats, bring the file to condition manually, and think one step ahead, for what the logo will be needed in six months.
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
